feat: handle trailing comments at EOF and complete EOF tolerance

[geek5nan]

Summary

This PR addresses two related issues with files that don't end with a newline:

1. Trailing comments at EOF produce error tokens

The lex() function's final_eol mechanism (line 306) indicates the intent to support files without trailing newlines — it injects a synthetic Eol when the token stream ends. However, when the last line contains an inline comment (; ...) without a following \n, the ; cannot match Eol (which requires \n) and becomes an error token, defeating final_eol.

Fix: Add an EolComment token (#[regex(r";[^\r\n]*")]) that matches trailing comments at EOF. Logos longest-match ensures Eol still wins when \n is present. EolIndentHandler converts EolComment to a zero-width Eol before it reaches the parser, maintaining full transparency.

2. Complete EOF tolerance in parsers (extends #54)

PR #54 added choice((Eol, end())) at 4 structural points. This PR extends the same pattern to the remaining 11 directive types (balance, price, open, close, event, note, document, pad, query, custom, commodity) for consistent behavior across all directives.

3. Harden resolve_span against invalid byte ranges

resolve_span used direct slice indexing (source_content[byte_span.start..byte_span.end]) which can panic on invalid byte boundaries. Changed to .get().unwrap_or("") for safety.

Motivation

• The final_eol mechanism already demonstrates the intent to handle missing trailing newlines
• The Beancount Python parser accepts all these cases
• 5 test cases previously marked as ANOMALY now pass correctly

Test plan

• All 335 existing tests pass
• 5 ANOMALY test cases (CommentEOF, ExtraWhitespaceComment, ExtraWhitespaceCommentIndented, ExtraWhitespaceTransaction, NoFinalNewline) now parse successfully
• Verified with real-world ledger files containing inline comments and missing trailing newlines

[tesujimath]

This is great, thanks!