fix(core): add pricing for current Claude models

[asjad3]

Description

calculateAnthropicCost matches the model against a prefix table and returns null when
nothing matches, so any model missing from the table shows no cost at all in the Console's
usage panel and in the CLI's per-request summary.

The table stops at Opus 4.6 / Sonnet 4.6, so most of the Claude models people are actually
running today fall through it:

Model	Before	After
claude-opus-5	no cost shown	$5 / $25
claude-opus-4-8	no cost shown	$5 / $25
claude-opus-4-7	no cost shown	$5 / $25
claude-sonnet-5	no cost shown	$3 / $15
claude-sonnet-4-5	no cost shown	$3 / $15
claude-haiku-4-5	no cost shown	$1 / $5
claude-fable-5	no cost shown	$10 / $50

Sonnet 4.5 and Haiku 4.5 are the ones I'd expect to bite people most — Opus 4.5 is in the
table but the two Claude 4.5 models next to it never were.

Cache-write and cache-read rates follow the same 1.25x / 0.1x relationship to input price
that every existing row uses.

Also relabelled three comments that named the wrong model (claude-3-opus was commented
"Claude Opus 4 (legacy)", claude-3-5-sonnet was "Claude Sonnet 4"). Comments only, no
behaviour change.

Prices are Anthropic's list prices. Sonnet 5 currently has promotional input/output pricing
running to 2026-08-31; I used the list price rather than the promo, since the table has no
way to express a time-boxed rate and the promo expires shortly.

Checklist

• I've read the contributing guide
• The relevant docs, if any, have been updated or created
• The relevant tests, if any, have been updated or created

Screen recording or screenshot

Not applicable — no UI change. The affected surfaces (TotalUsage.tsx, useLLMSummary,
useTotalUsage, and the CLI's streamChatResponse.helpers) all just consume the return
value of calculateRequestCost, which was previously null for these models.

Tests

Added 8 cases to calculateRequestCost.vitest.ts covering each newly-priced model, plus a
combined cache-read + cache-write case for Opus 5.

core/llm/utils/calculateRequestCost.vitest.ts   30 passed

With the test file kept and the change to calculateRequestCost.ts stashed, exactly the 8
new cases fail (8 failed | 22 passed), so they're pinning the fix rather than passing
either way.

The existing claude-unknown-model -> null case still passes; none of the added prefixes
match it.

---

Written with AI assistance (Claude Code); I reviewed the change and ran the tests above.

[github-actions]

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

[asjad3]

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