feat: harvest meaningful search categories for autocomplete

[EliezerIsrael]

Replace AutoCompleter._get_main_categories with _get_search_categories, which builds the category feed for autocomplete from the TOC tree.

• Keep the base-text browse taxonomy (categories with >= 2 children), recursing to full depth instead of stopping at two levels.
• At commentary/era/"Other" boundaries ("Commentary", "Rishonim on X", "Acharonim", "Modern", "Other ... Works", "Targum", "Guides"): drop the boundary node but harvest the good names one level below it (Rashi, Ramban, Kessef Mishneh, ...), without descending into their repeated per-book structure.

This removes the tens of duplicated commentary sub-trees (e.g. a full "Mishneh Torah" structure repeated under each of its commentators) while surfacing the commentator/author names that are valuable as search terms.

Description

A brief description of the PR

Code Changes

The following changes were made to the files below

Notes

Any additional notes go here

[gitvelocity-reviewer]

📊 Code Quality Score: 7/100

29 × 0.25 = 7.25, rounded to 7

Category	Score	Factors
🔭 Scope	5/20	Single file, single subsystem (autocomplete), localized change to one method and module-level constants
🏗️ Architecture	6/20	Replaces hardcoded 2-level loop with recursive walk; introduces boundary abstraction (constants + helper function); imports TocCategory for isinstance checks
⚙️ Implementation	8/20	Recursive tree traversal with closure mutation, regex + set boundary detection, meaningful behavior change from flat to recursive walk, isinstance guards
⚠️ Risk	7/20	Silently changes which categories appear in autocomplete search results; no feature flag; potential over-matching with 'other' keyword; unbounded recursion; no rollback mechanism
✅ Quality	2/15	Good docstring on _get_search_categories; is_category_boundary is well-factored; no tests added for new logic or regression coverage
🔒 Perf / Security	1/5	No recursion depth guard; otherwise low performance/security risk for this use case

Was this score accurate? 👍 Yes · 👎 No

_{Scored by GitVelocity · How are scores calculated?}