feat(providers): add Claude Sonnet 5.5 and Fable 5.1

[chuks-qua]

Summary

Adds Claude Sonnet 5.5 and Fable 5.1 to the Claude model catalog. Anthropic's effort docs list xhigh for every Claude 5 model, so Fable 5 and Sonnet 5 gain xhigh too.

 CLAUDE_STATIC_MODELS            effort ladder              context
   claude-opus-5-5               low..xhigh..max            1M
   claude-opus-5                 low..xhigh..max            1M
+  claude-sonnet-5-5             low..xhigh..max            1M
+  claude-fable-5-1              low..xhigh..max            1M
-  claude-fable-5                low, medium, high, max     1M
+  claude-fable-5                low..xhigh..max            1M
-  claude-sonnet-5               low, medium, high, max     1M
+  claude-sonnet-5               low..xhigh..max            1M

The shared effort and context-window tables (packages/shared) list the same IDs. Without that, a picked xhigh would drop to high before it reaches the SDK. The settings help text and docs/settings/reference.md list the same models.

Evidence

All captures come from an owned Electron process running this worktree's runtime. The before captures use main's sources.

• Before: the Claude picker has no Sonnet 5.5 or Fable 5.1.
  
  After: both appear with 1M context.
  

• Before: Fable 5's effort menu stops at Max with no X-High.
  
  After: Sonnet 5.5 offers Low through X-High and Max. Fable 5.1 and Fable 5 show the same menu.
  

• Server RPC provider.listModels { providerId: "claude" }:

  before: claude-opus-5-5, claude-opus-5, claude-fable-5, claude-sonnet-5, ...
  after:  claude-opus-5-5, claude-opus-5, claude-sonnet-5-5, claude-fable-5-1, claude-fable-5, claude-sonnet-5, ...
  PASS claude-sonnet-5-5 / claude-fable-5-1 / claude-fable-5 / claude-sonnet-5
       efforts=[low,medium,high,xhigh,max] ctx=1000000
  

• Tests:

  normalizeReasoningLevelForModel("claude-sonnet-5-5", "xhigh") -> "xhigh"   (was "high")
  normalizeReasoningLevelForModel("claude-fable-5-1-20270101", "xhigh") -> "xhigh"   (dated ID not shadowed by claude-fable-5)
  

  The focused runs pass:

  • packages/shared effort and context tests: 88 passed.
  • packages/contracts: 435 passed.
  • packages/providers Claude tests: 158 passed.
  • apps/web model registry, picker, settings and composer tests: 143 passed.
  • bun run typecheck and bun run lint are clean.

Merge Danger

Door: two-way

Static catalog entries and capability tables only. There is no migration and no persisted-data change.

Blast Radius: small

Fable 5 and Sonnet 5 users can now pick X-High, and it reaches the SDK instead of being lowered to high. If the API ever rejected xhigh for these models, turns at that level would fail. The docs say it is supported. Opus 5.5's default effort (docs say medium, the catalog keeps high) is out of scope.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}