feat(app): add Opus 5.5 to the Claude model picker

[chphch]

Claude Opus 5.5 shipped on 22 September and the Claude picker still stops at Opus 5, so reaching the new model from Happy means leaving the app and passing --model on the machine running the session. This adds two rows — claude-opus-5-5 and the 1M-context claude-opus-5-5[1m] — at the head of the Opus group. Every existing row is untouched, and so is the code default (claude-opus-5).

Why these exact keys

• Full IDs, not short aliases — the reason the list already documents in-file: the aliases do not all resolve to the model the row names.
• A separate [1m] row, following Opus 5's precedent rather than Fable 5.1's. The bracket suffix is part of the ID Claude Code accepts rather than a flag it drops, so the 1M build is a real second variant. Fable 5.1 has no such row because 1M is that model's default; Opus 5.5 has both.
• Both spellings were run against the installed CLI (2.1.280) before being offered here — claude --model claude-opus-5-5 and claude --model 'claude-opus-5-5[1m]' each returned a completion — so neither row can hand a session an ID the harness rejects.

Placement

At the head of the Opus group, not the top of the list. The catalog groups by family with each family's newest member first, and this keeps that. Opus 5.5 [1M] therefore sits between Opus 5.5 and Opus 5, matching how the Opus 5 pair is already ordered.

Proof

Screenshot in the comment below: the /new model picker open, showing all seven Anthropic rows in one frame with Opus 5.5 and Opus 5.5 [1M] above the currently-selected Opus 5. Captured on a throwaway standalone happy-server (PGlite) + Expo web, authenticated via the dev-token URL bypass and driven headlessly with Playwright; the row order was also read back from the live DOM rather than eyeballed. vitest run sources/components/modelModeOptions.test.ts passes with the assertions updated alongside, and tsc --noEmit is clean.

🤖 Generated with Claude Code

[chphch]

Proof — the picker offers Opus 5.5

Captured locally: a throwaway standalone happy-server (PGlite, port 3222) + Expo web (port 8252), authenticated through the dev_token URL bypass and driven headlessly with Playwright, booted from this PR's commit.

The row order was read back from the live DOM as well as screenshotted, so the frame is not the only evidence:

["Fable 5.1","Fable 5","Opus 5.5","Opus 5.5 [1M]","Opus 5","Opus 5 [1M]","Sonnet 5"]

Opus 5 carries the filled check because it is still the code default — this PR adds rows, it does not change what a new session starts on.

One framing note: the capture uses a 1000px-wide viewport on purpose. At ≥1100px the picker renders embedded and scrolls at ~4 rows (embeddedOptionList's maxHeight: 176), so the seven rows are all present but cannot share one frame. Below the sidebar threshold it renders as a popover and they do.