Syncs the models docs page with what `crates/billing/src/plans.rs` and
`crates/llm_api/src/features/list_models.rs` actually serve.

## Changes

**New models added:**
- GPT-5.4 pro (Pro/Business only): $30/$180 input/output, no cache, 400k
context
- GPT-5.4: $2.50/$15 input/output, $0.025 cached, 400k context
- GPT-5.3-Codex: $1.75/$14 input/output, $0.175 cached, 400k context

**Pricing corrected:**
- GPT-5.2 + GPT-5.2-Codex: input $1.25 → $1.75, output $10 → $14, cached
$0.125 → $0.175
- Gemini 3 Flash: input $0.30 → $0.50, output $2.50 → $3.00

**Other fixes:**
- Student plan note updated to include GPT-5.4 pro restriction
- Grok context windows added to the context window table (128k / 256k
for Grok Code Fast 1)
- "GPT-5.2 Codex" renamed to "GPT-5.2-Codex" to match `display_name` in
code
- Removed retired "Gemini 3 Pro" from context window footnote

Release Notes:

- N/A