v0.1a.2: tolerate PDF spacing differences in reference lookup

[drdeepaksatheesh]

Fixes #14.

The Windows smoke test showed that PDF.js can extract visually separated word runs as joined text (a given -> agiven). The v0.1a reference engine previously used exact normalized-text lookup and therefore rejected a supported passage.

Fix

• keep SourceAnchor text and quote hashing unchanged for provenance fidelity;
• introduce a reference-engine-only lookup key that removes punctuation/whitespace while preserving ordered alphanumeric content;
• add a regression test reproducing the exact agiven extraction seen in the smoke test;
• bump local reference engine to 0.1a.2 and app package to 0.1.0-alpha.2.

This is deterministic normalization, not semantic/fuzzy matching: unknown content must still return unsupported.

CI evidence

GitHub Actions run 34926919222: PASS.

• frontend/version checks: PASS
• unit tests including the agiven regression: PASS
• production build: PASS
• critical dependency audit: PASS
• Windows NSIS build: PASS
• installer validation + checksum: PASS
• artifact upload: PASS

Merged after CI passed.