fix(meta_analyzer): store None-end fallback key so static findings are not dropped

[Shrotriya-lalit]

Problem

LLMMetaAnalyzer.apply_filter() stored LLM-confirmed findings keyed by:

(file, rule_id, start_line, end_line)

Static analyzers almost never populate end_line — it defaults to None.
The LLM, however, fills in end_line for every finding it confirms.

This key mismatch meant every static finding with end_line=None was silently
dropped after LLM enrichment, even when the LLM had correctly confirmed it.
In practice, the meta-analyzer produced zero filtered findings for entire files.

Root Cause

# Before fix — only exact key stored:
confirmed_granular[(file, rule_id, int(start), int(end))] = enrichment
# Static finding has end_line=None → no match → finding dropped

Fix

When storing a granular LLM confirmation with an explicit end_line, also store
a fallback key with end_line=None via setdefault:

confirmed_granular[(file, rule_id, int(start), exact_end)] = enrichment
confirmed_granular.setdefault(
    (file, rule_id, int(start), None), enrichment
)

setdefault is used so that if the LLM also reports a finding at the same
start_line with end_line=None, its enrichment takes precedence. Lookup order
is unchanged: exact key → start-only key → coarse key.

Tests

• test_static_end_line_none_confirmed_when_llm_returns_end_line — reproduces the bug exactly
• test_static_end_line_none_multiple_same_rule — multiple findings same rule, different start lines
• test_end_line_used_when_provided — regression: exact end_line match still works

Closes #67

Checklist

• make lint passes
• Tests added for new behaviour (including bug reproduction test)
• DCO sign-off on commit (git commit -s)