feat: sync with tantivy-py 0.26.2

[oxdev03]

Update the tantivy-py submodule from 0.24.0-22 to 0.26.2 and port every upstream change back, keeping the file-for-file, method-for-method mapping the port is built on. Bumps tantivy 0.25 -> 0.26.2 and the napi-rs JS toolchain.

Ported APIs:

• Searcher: orderByField over u64/i64/f64/bool/date/str fast fields, weightByField, cardinality(), fastFieldValues(), termsWithPrefix(); aggregate() now takes and returns objects instead of JSON strings
• Query: andMustMatch/orShouldMatch/andMustNotMatch with flat chaining, booleanQuery minimumNumberShouldMatch, moreLikeThisDocumentFieldsQuery, unbounded range bounds, useInvertedIndex, offset pairs for phrase words, existsQuery(fastFieldName, jsonSubpaths)
• Index: isCompatible(), registerFastFieldTokenizer(), conjunctionByDefault and allowRegexes on both parse methods, deleteDocuments alias
• Module-level parseQuery/parseQueryLenient, addJsonField expandDotsEnabled

Fixes found while aligning:

• garbageCollectFiles was a silent no-op
• makeTerm truncated u64 values through get_uint32
• aggregate coerced its argument to "[object Object]"
• dates lost sub-second precision
• a negative value for an unsigned field silently became its absolute value
• parseQueryLenient returned formatted strings while the 17 error classes it should have returned were exported but unreachable
• .husky/pre-commit invoked yarn in a pnpm project, so it always failed

Deduplication: one term_from_value instead of two copies of the type match, one more_like_this_builder, get_field instead of four inlined lookups.

FieldType variants now use tantivy-py's names (Text, Unsigned, Integer, Float, Boolean, Json).

Docs: four new tutorial sections, each backed by an executable example.

BREAKING CHANGE: existsQuery drops its schema argument; aggregate takes an object; parseQueryLenient returns error instances rather than strings; phrasePrefixQuery and regexPhraseQuery drop maxExpansions and the two-term minimum; FieldType variants renamed; SearchHit.order widened to number | boolean | string; parser error messages now quote field names with backticks.