perf(dicer): avoid regex when parsing headers

[mcollina]

Summary

• avoid regex-based header field parsing in Dicer's HeaderParser
• scan CRLF-delimited header lines directly instead of using buffer.split(/\r\n/g)
• preserve folded-header handling and existing malformed-header behavior

Performance

Local HeaderParser microbenchmark on a 3-header block:

• before: ~1.78M ops/sec
• after: ~2.66M ops/sec

A local @platformatic/flame run also confirmed HeaderParser._parseHeader remains a primary hotspot, but its self-time dropped in the targeted multipart workload.

Validation

• npm test
• npm run lint

[ivan-tymoshenko]

lgtm

[marcopiraccini]

lgtm

[mcollina]

Added a follow-up optimization in 6d81635: HeaderParser now scans for the header terminator directly instead of routing header parsing through StreamSearch. I also added coverage for CRLFCRLF split across chunks, including after maxHeaderSize is reached.\n\nValidation:\n- npm test\n- npm run lint\n- npm run bench:dicer\n- remote CI checks are passing.

[mcollina]

Added another follow-up optimization in 978c38d, based on a full Busboy multipart flamegraph (Busboy -> Multipart -> Dicer -> HeaderParser), not just the Dicer micro/profile path.\n\nFull-profile result on the local multipart workload:\n- before this follow-up: 7,794 parses / 7s baseline from the full Busboy flamegraph\n- after HeaderParser direct scan: 8,610 parses / 7s\n- after parse/decode fast paths: 10,397 parses / 7s\n\nChanges in the latest commit:\n- skip Buffer/TextDecoder-style UTF-8 decoding for ASCII strings\n- add fast paths for common form-data; name=...; filename=... parsing\n- avoid parseParams() for simple per-part content-type values without parameters\n\nFinal full flamegraph artifacts are local at:\n- /tmp/busboy-full-flame-final/cpu-profile-2026-05-23T10-15-52-214Z.html\n- /tmp/busboy-full-flame-final/cpu-profile-2026-05-23T10-15-52-214Z.md\n\nValidation:\n- npm test\n- npm run lint\n- remote CI checks are passing.