fix(optimizer): preserve served model snapshots

[drewstone]

Problem

The official optimizer proxy required the execution receipt and response model to equal the requested alias byte-for-byte. A real Tangle Router call requested deepseek-v4-flash and returned the same model as deepseek/deepseek-v4-flash@fp_…; Runtime accepted that identity, but Eval rejected every optimizer proposal after the paid call.

Change

• reuse Eval's canonical served-model identity rule for optimizer callback receipts and responses
• preserve the provider-qualified snapshot in the response and cost ledger
• reject true substitution, unreported identity, and response/receipt disagreement
• release npm and Python packages as 0.145.3

Proof

• realistic pre-fix smoke: 7 optimizer callbacks, 6 rejected on the qualified snapshot mismatch and 1 bridge abort
• focused proxy tests: 46/46
• full suite before digest refresh: 5,192 passed, 3 skipped, with only the expected source-digest freshness test failing
• source and dependency digests refreshed and validated
• lint and typecheck pass
• merge-tree against current main succeeds

[tangletools]

✅ Auto-approved drewstone PR — d38edf08

This PR was opened by the trusted drewstone account.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: drewstone_author · 2026-08-12T16:49:15Z}