refactor(maven): modularize parser internals

[abraemer]

Summary

• replace the monolithic Maven parser with focused submodules for POM parsing, manifests, properties, shared coordinates, and split test files
• split POM state by domain, restore templated collection metadata resolution that was lost during the refactor, and extract shared META-INF/maven/... plus Maven URL/purl helpers
• narrow pom/state module boundaries so context.rs stays transition-focused and state/mod.rs relies on smaller intent-level child APIs

Scope and exclusions

• Included: structural Maven parser refactors, the missing templated repository/plugin-repository/mailing-list resolution fix, shared coordinate/URL helper extraction, and import-boundary cleanup
• Explicit exclusions: no new parser surfaces, no intentional output-shape changes, and no follow-up enhancement yet for developer/contributor party-field property resolution

Intentional differences from Python

• None intended beyond restoring the pre-refactor templated collection metadata resolution behavior that regressed on this branch.

Follow-up work

• Created or intentionally deferred: developer/contributor property resolution remains a separate follow-up if we want to expand Maven property substitution coverage further.

Validation

• cargo test maven::pom_test::tests:: --quiet
• cargo test maven::manifest_test:: --quiet
• cargo test maven::properties_test:: --quiet
• cargo test maven::scan_test::tests:: --quiet
• cargo test --features golden-tests maven_golden --quiet
• cargo test --features golden-tests test_assembly_maven_basic --quiet
• cargo test --features golden-tests test_assembly_maven_meta_inf_basic --quiet
• real-target metadata-only benchmark and result comparison against current main on apache/felix-dev @ 20aee77 covering pom.xml, *.pom, pom.properties, and MANIFEST.MF
• semantic diff on that target was empty after normalizing run-specific package_uid UUIDs; package counts matched exactly at 196
• performance on that target was effectively neutral versus main (branch 12.167s vs main 12.328s uncached-cold, identical 12.155s/12.157s uncached-repeat, identical warm incremental behavior)