feat(frontend): add native Kimi K3 request processing

[furionw]

Summary

• upgrade the released frontend crates for Kimi K3 rendering, parsing, and segmented tokenization
• add native K3 thinking, XTML tool-call, named-tool, and response wrapper handling
• preserve special-token boundaries through Rust, C, and inference-gateway preprocessing
• return client errors for invalid renderer inputs and malformed tool schemas

Stack

Base: main

PR2 and PR3 branch independently from this PR.

Validation

• cargo fmt --all -- --check
• cargo check -p dynamo-llm --no-default-features --locked
• targeted Kimi K3 tests: 4 passed
• pre-commit run --all-files
• Graham-style Rust review: PASS, no findings

Summary by CodeRabbit

• New Features
  • Improved Kimi K3 tool-calling support, including native XTML structural tags and better reasoning behavior.
  • Added model-aware prompt formatting and more accurate tokenization for rendered chat prompts.
  • Improved handling and reporting of invalid prompt-rendering requests.
• Bug Fixes
  • Corrected tool-choice validation and guided decoding behavior for Kimi K3 requests.
  • Validated tool parameter schemas more reliably while allowing omitted parameters.
• Documentation
  • Clarified command-line help and tool-choice guidance for Kimi K3 behavior.

[coderabbitai]

Walkthrough

The PR updates renderer dependencies, preserves rendered prompt segments during tokenization, adds Kimi K3 formatter and parser-specific reasoning/tool handling, validates tool parameter schemas, and updates integrations, help text, and tests.

Changes

Preprocessing and validation

Layer / File(s)	Summary
Renderer dependencies and Kimi K3 formatter selection Cargo.toml, lib/bindings/python/Cargo.toml, lib/llm/src/preprocessor/prompt.rs	Renderer and parser versions are updated, and Kimi K3 formatter selection is added before existing fallback formatters.
Rendered prompt tokenization and error mapping lib/llm/src/preprocessor.rs, deploy/inference-gateway/ext-proc/src/epp.rs, lib/bindings/c/src/lib.rs	Template rendering returns RenderedPrompt, maps invalid render requests, and uses segment-aware tokenization across preprocessing integrations.
Kimi K3 tool choice and structural tagging lib/llm/src/preprocessor.rs, lib/llm/src/preprocessor/structural_tag.rs, lib/llm/src/preprocessor/tool_choice.rs, components/src/dynamo/common/configuration/groups/runtime_args.py	Kimi K3 named-tool requests disable template thinking, require intrinsic XTML structural tags, and bypass incompatible JSON guided decoding.
Kimi K3 streaming and reasoning parser behavior lib/llm/src/preprocessor.rs	Kimi K3 aliases are included in tool-jail, special-token, reasoning-marker, and reasoning-gating paths with corresponding tests.
Tool parameter schema validation lib/llm/src/protocols/openai/validate.rs, lib/llm/src/protocols/openai/chat_completions.rs	Provided tool parameters must be JSON Schema objects; omitted parameters remain accepted.

Estimated code review effort: 4 (Complex) | ~45 minutes

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The description covers summary, stack, and validation, but it misses the required Overview, Details, reviewer-start, and Related Issues sections.	Rewrite it to follow the template: add Overview, Details, reviewer-start files, and the required Related Issues section with either Closes/Relates to #XXXX or a no-issue declaration.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title matches the main change: adding native Kimi K3 request processing, though it omits the broader preprocessing and schema-validation updates.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches 💡 1

🛠️ Fix failing CI checks 💡

• Fix failing CI checks

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

🧹 Nitpick comments (2)

lib/bindings/python/Cargo.toml (1)

56-56: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Same unpinned dynamo-parsers version as the root manifest.

See the consolidated comment covering both Cargo.toml and this file.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/bindings/python/Cargo.toml` at line 56, Update the dynamo-parsers
dependency declaration in the Python bindings manifest to use the same pinned
version specification as the root Cargo manifest, keeping both manifests
synchronized.

Cargo.toml (1)

52-62: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Inconsistent version pinning for dynamo-parsers.

dynamo-tokenizers, dynamo-renderer, and dynamo-protocols are pinned exactly (=1.7.0, =5.0.0, =5.0.1), but dynamo-parsers uses "7.0.0" (Cargo's default caret range, allowing >=7.0.0, <8.0.0). Since this PR is bumping several of these lock-step "frontend crates" together specifically to add K3 support, leaving one of them unpinned risks it drifting to an incompatible minor version relative to its siblings.

🔧 Proposed fix

-dynamo-parsers = { version = "7.0.0" }
+dynamo-parsers = { version = "=7.0.0" }

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@Cargo.toml` around lines 52 - 62, Pin the dynamo-parsers dependency exactly
to version 7.0.0, matching the exact-version constraints used by
dynamo-tokenizers, dynamo-renderer, and dynamo-protocols. Update only the
version declaration for dynamo-parsers and preserve its existing dependency
configuration.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@Cargo.toml`:
- Around line 52-62: Pin the dynamo-parsers dependency exactly to version 7.0.0,
matching the exact-version constraints used by dynamo-tokenizers,
dynamo-renderer, and dynamo-protocols. Update only the version declaration for
dynamo-parsers and preserve its existing dependency configuration.

In `@lib/bindings/python/Cargo.toml`:
- Line 56: Update the dynamo-parsers dependency declaration in the Python
bindings manifest to use the same pinned version specification as the root Cargo
manifest, keeping both manifests synchronized.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: fd3ea7fd-f331-462e-bb62-e6c2e973a090

📥 Commits

Reviewing files that changed from the base of the PR and between 0e1d1d8 and ce4c8f0.

⛔ Files ignored due to path filters (3)

• Cargo.lock is excluded by !**/*.lock
• lib/bindings/kvbm/Cargo.lock is excluded by !**/*.lock
• lib/bindings/python/Cargo.lock is excluded by !**/*.lock

📒 Files selected for processing (11)

• Cargo.toml
• components/src/dynamo/common/configuration/groups/runtime_args.py
• deploy/inference-gateway/ext-proc/src/epp.rs
• lib/bindings/c/src/lib.rs
• lib/bindings/python/Cargo.toml
• lib/llm/src/preprocessor.rs
• lib/llm/src/preprocessor/prompt.rs
• lib/llm/src/preprocessor/structural_tag.rs
• lib/llm/src/preprocessor/tool_choice.rs
• lib/llm/src/protocols/openai/chat_completions.rs
• lib/llm/src/protocols/openai/validate.rs

[rmccorm4]

#12203 (just merged) should help with TRTLLM CI test failures

  [w5]         error_data = {'code': 400, 'message': 'ValueError: node_id must be in range [0, 256)', 'type': 'Bad Request'}