feat(models): add LineNumber newtype to replace raw usize for start_line/end_line

[abraemer]

Summary

• Introduce LineNumber (backed by NonZeroUsize) and LineSpan newtypes to replace raw usize fields for start_line/end_line across the entire codebase
• Eliminate the possibility of invalid zero-valued line numbers at the type level, removing runtime guards that checked for start_line > 0
• Make the 1-indexed semantics of line numbers explicit in the type system

Key changes

• New src/models/line_number.rs with LineNumber(NonZeroUsize) and LineSpan { start, end } types
• All output structs (Match, Copyright, Holder, Author, OutputEmail, OutputURL) now use LineNumber for start_line/end_line
• All internal detection types (CopyrightDetection, HolderDetection, AuthorDetection, Token, EmailDetection, UrlDetection, LicenseMatch, FileRegion, DeclaredLicenseMatchMetadata) migrated
• SerializableLicenseMatch preserves JSON backward compatibility via serde(transparent) on LineNumber
• Removed now-unnecessary start_line > 0 / end_line > 0 runtime guards
• line_for_pos() returns Option<LineNumber> instead of Option<usize>
• matched_text_from_text() takes (LineNumber, LineNumber) params
• ranges_overlap() takes LineNumber params (Ord-based comparison)
• LineSpan defined but not yet adopted (for future use)
• LicenseMatch::default() uses LineNumber::ONE instead of 0

Performance

Neutral — NonZeroUsize has the same layout as usize, and Option<LineNumber> benefits from niche optimization (same size as Option<usize>).

Testing

• cargo build — 0 errors
• cargo fmt — clean
• cargo clippy --all-targets --all-features — 0 errors
• Targeted tests pass (license detection models, finder, debian parser, line-number specific tests)
• 1 pre-existing test failure unrelated to this change

How I tested

Ran targeted test suites for copyright detection, license detection models, finder, and debian parser modules. All pass except one pre-existing failure in test_normalize_company_suffix_period_holder_variants.