Read Many Followups

[tvaron3]

Description

Following two changes from suggestions from @analogrelay to read many/ query apis,

• Slight performance suggestion for a separate PR: Store request charges in an array with workerCount slots. Each worker can update the request charge slot using it's worker index w without a lock. Sum up all the slots at the end, when all the workers are done.

• As a refactoring for later, we can actually just collect all the results and call ProvideData once with a single batch. That's what it's designed for. You're right that it shouldn't be called simultaneously, but calling it once with a batch of items will also reduce CGo overhead.

[Copilot]

Pull request overview

This PR implements two performance optimizations to the runEngineRequests function based on code review feedback:

1. Lock-free charge tracking: Replaces mutex-protected charge accumulation with per-worker array slots, eliminating lock contention when workers record request charges.

2. Batched result delivery: Changes from calling ProvideData for each individual result to collecting all results and calling ProvideData once with a single batch, reducing CGo overhead.

Key Changes

• Removed the provider goroutine and chargeMu mutex
• Added per-worker charges array indexed by worker ID for lock-free charge accumulation
• Introduced a collector goroutine that gathers all QueryResult items into allResults
• Modified control flow to call pipeline.ProvideData once after all workers complete
• Changed function signature to use unnamed return values

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.