feat(catalog): add the current-generation models to the catalogue

[argszero]

Summary

The catalogue was missing every current-generation model the vendors shipped in October 2026, so a user could not select them on this platform at all: the [[models]] rows stopped at the previous flagships (zhipu glm-5.3, anthropic claude-opus-5, openai gpt-5.6-sol, google gemini-3.1-pro, aliyun qwen3.8-max, minimax minimax-m3). This PR adds the nine missing rows, additively — no existing row is touched.

New rows (prices verified against each vendor's official page on 2026-10-16; CNY per million tokens unless marked USD):

vendor	model	input	cache hit	output
zhipu	glm-5.3-flash	0.8	0.23	2.8
anthropic	claude-sonnet-5.5	$2.00	$0.10	$10.00
openai	gpt-6.1-sol	$4.00	$0.20	$20.00
openai	gpt-6-astra	$20.00	$2.00	$100.00
openai	gpt-6-luna	$0.20	$0.02	$1.00
google	gemini-3.8-flash	$0.75	$0.075	$3.75
google	gemini-4-argon	$2.00	$0.10	$10.00
aliyun-bailian	qwen3.8-max-prime	24	— (see below)	72
minimax	minimax-m2.7	2.1	0.42	8.4

Decisions a reviewer should know about:

• qwen3.8-max-prime cache-hit price — the official pricing table gives no figure for it; its row only says the model "enjoys a context-cache discount", exactly as the sibling qwen3.8-flash does. Recorded as 0 (cache hit free — it under-collects) with a comment marking it as pending, rather than copying another model's ratio or guessing one.
• minimax-m2.7 context window — neither the vendor's pricing page nor its model page publishes one. Recorded as context_length = 0 with a comment; no value was invented.
• Promotional prices carry an end date, and the config has no field for a date-triggered switch — gemini-3.8-flash ($0.75/$3.75 through 2026-12-31, then $1.50/$7.50) and gemini-4-argon ($2.00/$10.00 through 2026-11-30, then $4.00/$20.00) record the current value and spell out both figures plus the switch date in the row comment, so the next refresh has a source of truth.
• OpenAI service tiers — that pricing page has five (Standard / Batch / Flex / Fast / Ultrafast) and the config has no tier field, so each gpt-6 row records the Standard tier, with the tier note in the comment.
• glm-5.3-flash is marked vision = true — the vendor's page states native multimodality; every other new row is vision = false per its vendor's "image understanding" column.

Related Issue

Rant 2026-10-16T15:20:36.482297+08:00, keyword 【上架】 ("list these on the platform") — add the current-generation models to the catalogue. Its prerequisite, "make the four international vendors selectable when listing a key", landed as #414 (983111c); without it the anthropic / openai / google rows added here could not have been selected.

Changes

• config/config.example.toml — nine new [[models]] rows; the example file (which CI treats as the single source of truth) is updated in step with the change
• ui/js/data.js — the same nine models mirrored into MODELS (USD rows go through the existing USD() conversion)
• src/catalog_gate.rs — MODEL_COUNT 15 → 24
• ui/index.html — data.js?v= cache-bust token bumped

Tests

• cargo test — 564 passed, 0 failed
• cargo fmt --check — clean
• cargo clippy --all-targets -- -D warnings — clean
• New unit tests — not applicable: the change is catalogue data, and catalog_gate's three rules (bidirectional model-set equality between config.example.toml and data.js, plan-id set equality, MARKET ⊆ MODELS) are exactly the tests that cover it; MODEL_COUNT is their positive control.

Checklist

• Branch named per convention (feat/…)
• Commit message in Conventional Commits format
• Single responsibility, minimal change