Fix for long document load times introduced in v0.35.0

[mcantrell]

Summary

Removes nom_locate dependency to fix a performance regression introduced in v0.35.0. See #412 for details.

The parser input type was changed from &[u8] to LocatedSpan<&[u8], &str> in 0.35.0 to carry debug labels through the parser. LocatedSpan tracks line/column position by scanning for newlines (via memchr::count_raw) on every slice operation. For a large PDF, this means millions of redundant newline scans during parsing.

Note

Neither the line/column tracking nor the debug labels were ever read.** The only LocatedSpan methods actually used (.len(), .take_from(), .fragment()) all have direct equivalents on &[u8].

Fix

• Removed nom_locate from Cargo.toml
• Changed ParserInput<'a> from LocatedSpan<&[u8], &str> back to &[u8]
• Replaced all ParserInput::new_extra(bytes, "label") construction sites with plain byte slices
• Fixed minor type adjustments in trim_spaces signature, a verify closure, and cmap_parser test assertions

Result

Loading a 100-page PDF dropped from 9-10 seconds to ~10ms

Test:
tests/document_load_performance.rs with tests/regression/test.pdf

Warning

I've included the test.pd for testing but you may not actually want it. I just wanted to point it out because it's quite large. I can remove from the PR if you'd like.