fix(cost): resolve databricks_v2 pricing and surface cost in standard usage update

[jedwards27]

Fixes #9924.

Problem

Per-session cost (accumulated_cost) was effectively unavailable to ACP clients for the databricks_v2 provider:

1. Pricing never resolved. is_meta_provider() listed only "databricks", so databricks_v2 skipped meta-provider model inference. The provider's model aliases (e.g. goose-claude-opus-4-8) never mapped to a canonical model, so no price was found and accumulate_cost returned None.
2. Cost not on the wire clients read. Even once computed, cost was attached only to the custom goose notification. The standard ACP UsageUpdate — which clients consume, and which has a cost: { amount, currency } field — left cost: None.

Fix

• crates/goose-providers/src/canonical/name_builder.rs: add "databricks_v2" to is_meta_provider so its aliases resolve to canonical models and get priced (same path as legacy databricks).
• crates/goose/src/acp/server.rs (build_usage_updates): populate cost on the standard UsageUpdate from session.accumulated_cost via .cost(Cost::new(amount, "USD")).

Tests

• name_builder: databricks_v2 aliases resolve like the legacy databricks provider.
• server: the standard usage update serializes cost when present ({"used":...,"size":...,"cost":{"amount":...,"currency":"USD"}}).

Validation

• cargo test -p goose-providers --lib canonical::name_builder — pass
• cargo test -p goose --lib test_standard_usage_update_serializes_cost — pass
• cargo fmt --check and cargo clippy — clean on the changed crates

Notes

• Cost remains an estimate (model list prices via the canonical catalog), not a billed amount.
• I verified end-to-end with an A/B: a client rendering UsageUpdate.cost shows blank with a stock engine and a real cost figure with this fix; the only variable changed was the engine binary.

[jedwards27]

Pushed a fix for the failing CI checks. The Build/Test, MSRV, and Lint jobs were all failing for the same root cause — the test I added didn't compile because the branch was cut from an older base where the test helpers had a different shape.

Fixed by rebasing onto latest main and updating test_standard_usage_update_serializes_cost to match current conventions:

• make_session_with_usage(...) now takes two TokenUsage args (was 6 Options) — updated to make_session_with_usage(TokenUsage::new(Some(60), Some(40), Some(100)), TokenUsage::default()).
• ModelConfig path corrected to goose_providers::model::ModelConfig.

No change to the actual fix (databricks_v2 meta-provider + cost on the standard UsageUpdate). Locally: cargo fmt --check clean, cargo clippy clean on the touched crates, and the cost serialization test passes. Re-running CI now.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d8f0ad4105

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[DOsinga]

Thanks @jedwards27 — clean, well-scoped fix for a real gap: databricks_v2 was missing from is_meta_provider() so its aliases never resolved to canonical catalog models for pricing, and the cost wasn't surfaced on the standard ACP UsageUpdate.

I pushed a small cleanup commit: dropped the two added tests (they asserted pre-existing meta-provider inference behavior against hardcoded catalog versions rather than this change) and removed a redundant comment. The substantive change is the right one-liner plus the cost surfacing. Approving.

[jedwards27]

Thanks @jedwards27 — clean, well-scoped fix for a real gap: databricks_v2 was missing from is_meta_provider() so its aliases never resolved to canonical catalog models for pricing, and the cost wasn't surfaced on the standard ACP UsageUpdate.

I pushed a small cleanup commit: dropped the two added tests (they asserted pre-existing meta-provider inference behavior against hardcoded catalog versions rather than this change) and removed a redundant comment. The substantive change is the right one-liner plus the cost surfacing. Approving.

Thank you @DOsinga ! I resolved the bot's comment after you posted, let me know if you'd like me to wait for your re-review (otherwise I'll go ahead and merge this later today).

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1f8345ef31

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".