Fix whitespace-only header line silently truncating headers (fixes #222)

[scadastrangelove]

Fixes #222.

ParserConfig::allow_space_before_first_header_name(true) + a line consisting only of whitespace right
after the request/status-line silently dropped the entire header block: the parser returned
Ok(Status::Complete) with headers == [], leaving every real header (Host, Content-Length, ...)
unconsumed in the buffer.

The fix

After stripping the leading whitespace run, peek at the next byte before deciding to continue:

match bytes.peek() {
    None => return Ok(Status::Partial),
    Some(b'\r') | Some(b'\n') => break 'header Error::HeaderName,
    _ => {}
}

• A whitespace-only line (next byte is \r/\n) now rejects with Error::HeaderName — the same error
  the non-leniency path already returns for a malformed first byte, rather than being silently absorbed
  as the terminating blank line.
• If there's no more data yet (peek() returns None), returns Status::Partial so the streaming/
  incremental case is unaffected — a caller that hasn't yet read the byte deciding whitespace-only vs. a
  real header isn't penalized.
• The documented, tested, non-degenerate case (space followed by a real header) is untouched.

Tests added

• test_allow_space_before_first_header_name_rejects_whitespace_only_line — both request and response
  variants of the whitespace-only-line case now return Err(HeaderName) instead of silently truncating.
• test_allow_space_before_first_header_name_partial_after_whitespace — a buffer ending right after the
  whitespace run still returns Status::Partial, confirming the fix doesn't regress incremental parsing.

Validation

• Full test suite: 99 → 101 tests, all green (cargo test --release).
• Existing test_allow_response_response_with_space_before_first_header (the documented non-degenerate
  case) unaffected.
• Manually re-verified both PoCs from the issue now return Err(HeaderName) instead of
  Ok(Complete)/headers==[].

Found & fixed with the rust-in-peace security pipeline.