fix(fts): drop the tokenize macro with the index

[zahariash]

This work was produced with the help of language models.

Fixes #119.

DROP_FTS_INDEX left behind the index's tokenize macro. Recreating an index under the same name reused the old macro, so rows built by the CREATE kept the old tokenization while rows inserted later and queries used the new one. The other half of #119, the CREATE failing on a leftover stopwords copy, is fixed on main by #104.

Fix. DROP_FTS_INDEX drops the macro after the index tables and now returns Index <name> has been dropped.; the NOTE #104 left there is gone, since LadybugDB/ladybug#1165 fixed DROP MACRO's WAL replay (LadybugDB/ladybug#1163). CREATE_FTS_INDEX replaces an existing tokenize macro, including the one schema.cypher creates during IMPORT DATABASE and one left by an interrupted CREATE. The macro is dropped only when it exists, because DROP MACRO IF EXISTS throws for a missing macro. Rebased onto main after #104; this overlaps #125 in create_fts_index.cpp, and whichever lands second gets rebased.

Tests. drop_recreate.test recreates an index with a new ignore_pattern and over a leftover macro, and drops an index with a stopwords table, reopens without a checkpoint so the drop is replayed, and recreates it with a new ignore_pattern, which finds 12345; these three fail on main. Its stopwords table and stopwords file cases already pass on main since #104 and stay as regression tests. multi_index.test no longer drops the macros before IMPORT DATABASE. FTS extension tests: no new failures.