fix(core): price gpt-4.1 family at OpenAI list rates instead of legacy gpt-4

[weike-zhang]

Fixes #13184

Summary

calculateOpenAICost matches pricing rows by longest-prefix startsWith. The table had no gpt-4.1* rows, so every gpt-4.1 id (including date-stamped snapshots like gpt-4.1-2025-04-14) fell through to the legacy gpt-4 row (input $30 / output $60 per MTok) — 15x–300x over OpenAI list price.

What this does

Adds the three missing pricing rows:

model	input	output
gpt-4.1-mini	0.40	1.60
gpt-4.1-nano	0.10	0.40
gpt-4.1	2.00	8.00

(per million tokens, USD)

No matcher change needed: sortedKeys is longest-first, so gpt-4.1-mini/gpt-4.1-nano (12 chars) are checked before gpt-4.1 (7), which is checked before legacy gpt-4 (5).

Verification

Regression cases added to core/llm/utils/calculateRequestCost.vitest.ts (1000 in / 500 out tokens):

• Before the fix, all four new cases fail — every gpt-4.1* model cost 0.06 USD (legacy gpt-4 rates).
• After the fix, all pass:
  • gpt-4.1: 0.006, gpt-4.1-mini: 0.0012, gpt-4.1-nano: 0.0003, gpt-4.1-2025-04-14: 0.006
  • control gpt-4o unchanged at 0.0075

Prettier check clean on both files.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[weike-zhang]

I have read the CLA Document and I hereby sign the CLA

[weike-zhang]

Note: the two failing checks are infrastructure-related and pre-existing — jetbrains-tests dies extracting a corrupted xz artifact, and the e2e job fails in @vscode/ripgrep's postinstall (GitHub API 403 while fetching ripgrep-prebuilt). Both also fail on other open PRs (e.g. #13179) and neither touches this change (core/llm/utils/calculateRequestCost.ts + its test only).