Gate blank-typed link inference with a structural-home check

[kdroidFilter]

Problem

In the reader's commentator (מפרשים) panel, opening a commentary on a verse showed extra, unrelated segments. Example: Shulchan Arukh OC 323:6 → Magen Avraham listed 5 comments, but only ס"ק ח/ט (323:8, 323:9) actually comment on that se'if — the others (302:6, 321:7, 331:5) are in different simanim.

Root cause

Genuine commentary links are explicitly typed commentary in Sefaria; lateral cross-references have a blank Conection Type. The generator's blank-type inference (inferConnectionTypeFromSchema) promoted any blank-typed link to COMMENTARY whenever a base→dependant relationship existed, so cross-references (a commentary citing another siman) were mislabeled as commentary.

Fix

inferBlankConnectionType wraps the schema inference with a structural-home gate: keep the oriented promotion only when the dependant segment's top-level structural index (siman / perek) matches the base it points at; otherwise classify as REFERENCE. Daf-style / whole-book refs are left untouched (never demoted).

• Explicitly-typed commentary links are unaffected.
• Blank-but-home commentaries (e.g. Abarbanel → the verse it expounds) stay COMMENTARY.

Validation

Regenerated the DB seeded from the v1 build state:

• 323:6 → Magen Avraham: 5 COMMENTARY → 2 COMMENTARY (the genuine ס"ק ח/ט), 3 → REFERENCE.
• Corpus-wide COMMENTARY −1.35% (concentrated in halakhic cross-refs); Genesis 1:1 kept 70/72 commentators, the 2 dropped being real cross-refs to other parashot; Rashi intact.
• Unit tests added in SefariaLinksImporterTest.