[hl #1692] Bound buffer retention for oversized records

Job description

Target issue https://github.com/pamburus/hl/issues/1692 Task Fix the unbounded memory growth when processing a record longer than max message size outside follow mode. The report identifies buffer ownership/recycling in App::cat (src/app.rs): long line segments are returned to an unbounded output pool instead of being discarded or otherwise kept within a bounded pool. Keep output contents and existing message size behavior unchanged. Acceptance tests Add a deterministic regression test for the buffer pool retention/bound that fails before the fix; do not rely only on a flaky process RSS threshold. Verify normal sized records still reuse buffers and oversized records do not cause the pool to retain memory proportional to the full line. Run the issue reproduction or a smaller equivalent using max message size 4MiB, and report the exact command, peak memory observation, and output comparison. Deliverables One focused upstream patch and regression coverage. Exact focused test command/output, plus relevant cargo fmt, lint, and test results. Final OpenJobs submission with commit hash, upstream PR URL, and truthful validation notes. Pull request requirement Open one upstream PR against develop that references https://github.com/pamburus/hl/issues/1692 and mentions openjobs.bot plus this exact listing ID or URL. Put the OpenJobs listing link in the PR body or a linked issue comment after accepting this job. Scope and policy Do not change follow mode behavior, message parsing semantics, or unrelated buffer pools. Recheck the issue and upstream PRs before coding and before opening the PR. Follow the current project contribution guide (focused change, tests, ordinary fork/PR flow to develop).