perf(cli): buffer scan stdout writes

[mizchi]

Summary

• buffer scan stdout in 64 KiB chunks
• preserve the existing per-hit writer callback used by tests
• retain JSONL output and stderr streaming

Benchmark

Native release, 1,000 copies of matching/matching.mbt (~56,000 JSON hits):

version	mean
before	2.645 s
buffered stdout	2.316 s

This is a 12.5% improvement; system time fell from 0.373 s to 0.170 s.

Validation

• moon info && moon fmt
• moon check
• moon test (422 passed)
• git diff --check

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: fa596cefe3

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".

[chatgpt-codex-connector]

Flush sparse matches while scanning

When fewer than 64 KiB of matches have accumulated, this condition withholds every result until the entire traversal finishes. Thus an early match followed by a long scan produces no observable stdout for a pager or JSONL consumer, regressing the documented behavior that moongrep outputs results directly as it scans (SKILL.md:23-28) and the function's own streaming contract. Retain batching for throughput, but add a time-/idle-based flush or otherwise ensure sparse results become visible before completion.

Useful? React with 👍 / 👎.