refactor: semantic analyzer modules exceed 1000-line hard limit

[krystophny]

Summary

Two modules exceed the hard limit of 1000 lines per module (CLAUDE.md policy):

1. src/semantic/analyzers/semantic_analyzer.f90: 1360 lines (exceeds by 360 lines)
2. src/semantic/analyzers/semantic_function_analysis.f90: 1104 lines (exceeds by 104 lines)

Context

These files were already over the limit before recent PRs. PR #1922 added functionality but immediately extracted new code into semantic_call_signature_collector.f90 to minimize impact.

Proposed Solution

Refactor both modules by extracting cohesive functionality into separate modules:

For semantic_analyzer.f90:

• Extract statement inference logic (~300 lines)
• Extract context management utilities (~100 lines)
• Extract type finalization helpers (~50 lines)

For semantic_function_analysis.f90:

• Extract parameter analysis (~250 lines)
• Extract return type determination (~150 lines)
• Extract scope creation logic (~100 lines)

Acceptance Criteria

• Both modules under 1000 lines (preferably under 500)
• All tests pass
• No functionality changes
• Clear module boundaries and minimal coupling

Priority

polish - Technical debt cleanup, does not block functionality