feat(catalog): add openai/gpt-5.5 — flagship released 2026-04-23

[1bcMax]

Tracks BlockRunAI/blockrun#6, which added `openai/gpt-5.5` to the gateway catalog as the new OpenAI flagship (1M context / 128K output / $5/$30 per M tokens / fallback `gpt-5.4`).

Client-side wiring across the eight places Franklin keeps a hand-maintained model list:

File	Change
`src/pricing.ts`	New row: `$5` input / `$30` output
`src/agent/tokens.ts`	Conservative 128k context cap (matches every other gpt-5.x line; bump to the gateway's 1.05M is a separate change once verified end-to-end)
`src/agent/optimize.ts`	32k max-output cap, matching gpt-5.4
`src/agent/compact.ts`	Tier-down heuristic treats gpt-5.5 same as gpt-5.4 / sonnet (compaction model → haiku)
`src/ui/model-picker.ts`	`gpt` / `gpt5` / `gpt-5` aliases now resolve to gpt-5.5; `gpt-5.4` keeps its explicit alias. Picker list adds 5.5 as highlighted row
`src/proxy/server.ts`	Same alias updates as the picker
`src/router/index.ts`	AUTO MEDIUM/COMPLEX and PREMIUM COMPLEX fallback chains: `gpt-5.4` → `gpt-5.5`. Gateway falls back internally to 5.4 on 5.5 errors, so client chain stays the same length
`src/index.ts`	CLI `--model` help example updated

Test plan

• `npm run build` clean
• `npm test` — 167/167 pass (existing model-pinning tests still reference 5.4 explicitly, so they keep working as a regression check)
• Live: `franklin --model gpt -p "hello"` resolves to gpt-5.5 and the gateway accepts the call