v0.6.5: zero-copy string literals

v0.6.5: zero-copy string literals

v0.6.4: per-parse arena allocator (sync.Pool + slice arena)

v0.6.4: per-parse arena allocator (sync.Pool + slice arena)

v0.6.3: no-newline fast advance, comment IndexByte, sawNonASCII (#80)

Three per-byte redundancies removed from the lexer hot paths:

- skipSpace and the indent-block loop in nextInternal now use
  strings.IndexByte to jump past comment text in one SIMD call instead
  of advancing character by character.

- scanNameOrPrefixedString and the digit loops in scanNumber use
  s.off++; s.col++ directly for bytes that can never be newlines,
  skipping the '\n' branch inside scanner.advance.

- sawNonASCII flag set during identifier scanning replaces the second
  pass that hasNonASCII did over every identifier. hasNonASCII removed.

+12% serial throughput (92.4 MB/s vs 82.4 MB/s, 5x20s back-to-back).

v0.6.2: ASCII character classification lookup tables

v0.6.2: ASCII character classification lookup tables

v0.6.1: parallel file processing

v0.6.1: parallel file processing