Read turbopuffer query result ids from the raw response

[owenhalpert]

The turbopuffer client builds a pydantic model per returned row, but only the ids are used. Interleaved A/B on the same queries, the model construction is about 28% of client CPU per query at top_k=100 (1.30ms -> 0.94ms), rising to 47% at top_k=1000.

Reading the ids off the raw response instead: 2,910 -> 3,700 QPS on Cohere 1M (768d), recall unchanged at 0.9045. No new dependencies, loads is already imported.

Marking as draft while I confirm there are no edge cases with the multitenant and payload profile paths.

[sre-ci-robot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: owenhalpert
To complete the pull request process, please assign xuanyang-cn after the PR has been reviewed.
You can assign the PR to them by writing /assign @xuanyang-cn in a comment when ready.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[owenhalpert]

Closing this — my benchmark numbers were not measured rigorously enough. Re-running as a proper alternating A/B shows a much smaller difference than the PR claimed. Will revisit if it holds up.