perf: Optimize ingest delivery for logs

[Ankur Goyal (ankrgyl)]

We waste CPU in some places and can make ingest 2-3x cheaper.

Local pyperf benchmarks on CPython 3.14.7; times are means in microseconds.

Operation	Before (µs)	After (µs)	Speedup
Queue: 1,000 inserts	248.24	82.48	3.01×
Attachment discovery: small payload	1.69	0.96	1.77×
Attachment discovery: medium payload	14.95	7.62	1.96×
Attachment discovery: large payload	129.77	56.98	2.28×
Batching: 200 items, count limit	14.22	10.12	1.41×
Batching: 200 items, byte limit	21.49	12.61	1.70×
Batching: 200 items, both limits	22.20	14.06	1.58×

These measure SDK CPU overhead.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	⚠️ Failed 2026-10-05T06:20:49.426399Z	0001b69	PR opened

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 (@codex) review" or "Codex (@codex) security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[Abhijeet Prasad (AbhiPrasad)]

will take this PR over, have some small adjustments to make before merging

[Abhijeet Prasad (AbhiPrasad)]

released with https://github.com/braintrustdata/braintrust-sdk-python/releases/tag/py-sdk-v0.44.1