fix(acp): advertise model as SessionConfigOptionSelect on ACP 0.11+

[Christopher-Schulze]

What does this PR do?

Updates the Hermes ACP server to the agent-client-protocol >= 0.11 session-config-option surface. The model selector is now advertised as a SessionConfigOptionSelect in NewSessionResponse.configOptions / LoadSessionResponse.configOptions, and model changes are handled through session/set_config_option instead of the removed session/set_model method.

Related Issue

Fixes #80575

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• acp_adapter/server.py:
  • Replaced ModelInfo / SessionModelState / SetSessionModelResponse with SessionConfigOptionSelect / SessionConfigSelectOption and SetSessionConfigOptionSelectRequest semantics.
  • _build_model_state() now returns a SessionConfigOptionSelect (id="model", type="select", grouped by provider description).
  • new_session() / load_session() populate config_options with the model selector.
  • set_config_option() handles the "model" config ID by resolving and applying the chosen provider/model.
  • Added defensive type checks on state.model and state.agent.provider.
• pyproject.toml: relaxed agent-client-protocol==0.9.0 to >=0.11.0,<1 in the acp extra (upper bound required by project dependency policy).
• tools/lazy_deps.py: updated the tool.acp lazy pin from ==0.9.0 to ==0.11.0 so it matches the locked version.
• uv.lock: regenerated for the new ACP version.

How to Test

1. uv sync --extra all --extra dev
2. uv run python -m pytest tests/acp/test_server.py tests/acp/test_named_provider_catalogs.py -q
3. scripts/check.sh --project hermes-agent --worktree worktrees/hermes-agent/80575

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature
• I've added tests for my changes
• I've tested on my platform: macOS

[alt-glitch]

This was generated by AI during triage.

Related to #75358: both add ACP model configOptions, but this PR migrates fully to ACP 0.11+ while #75358 retains the legacy models surface for backwards compatibility.

[Christopher-Schulze]

Rebased this existing PR onto the current main, retained the current MCP 2.x pins, kept the ACP >=0.11 migration, and regenerated uv.lock. Focused ACP verification: 36 tests passed; uv lock check passed.

[Christopher-Schulze]

Fixed the CI metadata failure by aligning tools/lazy_deps.py with the locked agent-client-protocol==0.12.0. Verification: project metadata, ACP catalog, and ACP server tests pass (43 total). Updated head: 1b1117c785.

[Christopher-Schulze]

Rebased onto current main and resolved the merge conflicts.

acp_adapter/server.py keeps the ACP 0.11+ SessionConfigOptionSelect surface and ports the empty-catalog filter from main onto select_options (so an authoritative empty named catalog still does not resurrect the current model). The matching regression now asserts config_options[0].current_value == "".

[Christopher-Schulze]

Rebased fix/80575-acp-session-config-options onto current main (1a19fedb5a) so GitHub picks up a new head SHA (f2fbbc72b5). No code changes beyond the rebase; ACP SessionConfigOptionSelect + empty-catalog filter on select_options are unchanged.

[Christopher-Schulze]

Rebased onto current main (13ce0c5c67); new head 1bde66fa62. No code changes beyond the rebase — the diff still touches only ACP files (acp_adapter/server.py, pyproject.toml, tools/lazy_deps.py, uv.lock, tests/acp/*).

For context on the two red checks from the previous head: neither is caused by this branch.

• Python tests / Run tests slice 12/12 — test_upscale_defaults_are_all_off fails because ceabb030fb shipped the Grok Imagine Image 2.0 catalog entry with upscale: True, violating the opt-in-only upscaling policy from f06c41522e. main itself is red on this exact test (e.g. run 32240262381 @ 13ce0c5c67), so it fails on every PR until the catalog flag is flipped. Maintainer-side fixes are already open (fix(image_gen): Grok Imagine 2.0 upscale default no longer breaks main CI (opt-in policy) #89933, Main is green again: FAL Grok 2.0 upscale flag follows the opt-in policy #89935, fix(image-gen): Grok Imagine 2.0 no longer defaults upscale on — unbreaks main CI #89939, fix(image): keep grok upscale opt-in #89946).
• Python tests / e2e — test_plaintext_restart_gateway_in_group_stays_plain_text[telegram] is a flake; it passed on main in the same window (run 32240262381, 61 passed).

All ACP tests (tests/acp/*, tests/acp_adapter/*) and the other 10 slices passed on the previous head, and this rebase changes no code. Once the Grok upscale flag lands on main, this PR's CI should go green.