Entity Footnotes patch v1.0.4

Fixes scanning finding 0 matches. The v1.0.1 boundary-matching change (replacing \b with a hand-built boundary pattern) was based on an incorrect diagnosis - crengine's regex engine (SRELL) is a real ECMAScript-compliant library that supports \b correctly, confirmed by reading its C++ source. The replacement pattern most likely failed to compile, which makes the search silently return zero hits. Reverted to the original \b(...)\b matching.

Also adds diagnostic logging (term counts at scan start, and separate log lines for each init step) so if something still doesn't work, we can actually see why from the KOReader log instead of guessing again.

The localization-injection fix from v1.0.3 is unchanged and still needs verification - if strings are still showing as raw keys after this update, please check the KOReader log for lines starting with "EntityFootnotesPatch:" and share what they say.