Add Fireworks AI Provider

[alex-fireworks]

Summary

Adds Fireworks AI as a bundled, first-class BYOK model provider across the CLI, web dashboard, TUI, and desktop onboarding. Paste a FIREWORKS_API_KEY and go — models are addressed by their native catalog IDs (e.g. accounts/fireworks/models/kimi-k2p6).

This comes from Fireworks directly and consolidates the prior overlapping Fireworks PRs (supersedes #28983; also folds in the scope of #23541, #15332, #6054, #3757, #3564) into a single, mergeable change.

What's included

• Provider plugin (plugins/model-providers/fireworks/) — ProviderProfile with attribution headers (HTTP-Referer / X-Title), and PAYG-safe default aux + fallback models. Model IDs track the canonical Fireworks catalog (fw-ai/fireconnect).
• Picker registration — Fireworks shows up in hermes model, the web dashboard, and the TUI.
• Alias resolution — fireworks-ai / fw resolve through both CLI resolvers (hermes_cli/providers.py and hermes_cli/models.py), not just the plugin registry.
• First-class wiring — OPTIONAL_ENV_VARS, HERMES_OVERLAYS (with a FIREWORKS_BASE_URL override), and doctor env hints. The live catalog and model_metadata reverse-mapping are auto-derived from the profile.
• doctor — treats Fireworks' native slash-form IDs (accounts/fireworks/...) as valid rather than aggregator vendor prefixes, so it no longer suggests switching to openrouter / dropping the prefix.
• Picker default — plugin providers with no static curated list now lead with their profile fallback_models, so the default is an agentic chat model instead of whatever the live catalog returns first.
• Desktop onboarding — Fireworks as a RECOMMENDED hero card with the official Fireworks logomark, routing to the BYOK key form; copy localized in en/ja/zh/zh-hant.

Addresses prior review feedback (from #28983)

• The plugin-discovery test asserts the directory-count invariant (no hardcoded provider count); Fireworks is added to the representative spot-check.
• Alias resolution is wired into — and tested against — the CLI/TUI resolvers, not only the profile.
• The live smoke test drives the Hermes runtime (resolve_provider_client) rather than a raw OpenAI SDK client, so it exercises the real auth/config/runtime path.

Test plan

• pytest tests/plugins/model_providers/test_fireworks_profile.py tests/hermes_cli/test_fireworks_provider.py tests/providers/test_plugin_discovery.py
• Provider regression suite (parity / catalog / api-key / gmi) — no regressions
• Desktop tsc --noEmit + onboarding vitest
• Live: single-turn chat and attribution headers verified end-to-end against a Fireworks key
• HERMES_LIVE_TESTS=1 FIREWORKS_API_KEY=... pytest tests/run_agent/test_fireworks_live.py (opt-in)

Fire Pass (fpk_) support is coming soon; the future wiring is kept as a commented-out scaffold in the plugin.

[alt-glitch]

This was generated by AI during triage.

Related: this consolidates the earlier Fireworks provider attempts -- #6054 and #15332 (still open), plus (per the PR body) supersedes #28983 and folds in #23541 / #3757 / #3564. It comes from Fireworks directly and is the single mergeable version of the cluster, so the earlier community PRs can be closed in its favor once a maintainer confirms. Cross-linking the open siblings for navigation; not marking anything a strict duplicate.

[tonydwb]

Code Review Summary

Verdict: Comment (new provider + desktop UI)

This PR adds Fireworks AI as a preferred provider, touching 20 files including desktop onboarding UI, i18n, provider config, and tests. The scope is broad but additive.

Looks Good

• Clean provider integration pattern
• Desktop UI SVG assets included
• Multi-language support extended
• No security concerns

[teknium1]

Fireworks AI support landed via PR #62593. This contribution was part of the Fireworks provider cluster credited in the salvage PR; thank you for the implementation and review work.