… types

There was no system table exposing the available data skipping index types
(`system.data_skipping_indices` lists per-table index instances, not types).
This adds `system.data_skipping_index_types`, and attaches the shared
`Documentation` struct (introduced for table engines) to skipping index types.

- `MergeTreeIndexFactory::registerCreator` now takes a final `Documentation`
  argument, stored in a per-type map, with a `getDocumentation` accessor.
- The new `system.data_skipping_index_types` table exposes `name` and the
  embedded documentation columns `description`, `syntax`, `examples`,
  `introduced_in`, and `related`.
- All index types (`minmax`, `set`, `ngrambf_v1`, `tokenbf_v1`, `sparse_grams`,
  `bloom_filter`, `vector_similarity`, `text`, `hypothesis`) get a description
  and syntax.

This is a follow-up to the embedded-documentation changes for table engines,
database engines, data types, formats, dictionary layouts/sources, and
aggregate function combinators, and reuses the `Documentation` struct from
`src/Common/Documentation.h`.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Three of the embedded `syntax`/`description` strings for data skipping
index types described declarations that the validators would reject or
referenced functions that do not exist:

- `sparse_grams`: `bloomFilterIndexTextValidator` requires 5 or 6
  arguments (two or three tokenizer parameters before the bloom filter
  parameters), but the string showed a 4-argument form. Use
  `sparse_grams(min_ngram_length, max_ngram_length[, min_cutoff_length], size_in_bytes, num_hash_functions, seed)`.
- `vector_similarity`: `vectorSimilarityIndexValidator` requires the
  `dimensions` argument (3 or 6 arguments total), but the string omitted
  it. Add `dimensions` after `distance_function`.
- `text`: the description referenced `searchAny` and `searchAll`, which
  are not ClickHouse functions. Reference the actual text-search
  functions `hasToken`, `hasAnyTokens`, `hasAllTokens`, and `hasPhrase`.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Address review feedback on the regression for
`system.data_skipping_index_types`:
- Assert that no registered index type exposes an empty `description` or
  `syntax`, instead of checking only a handful of hard-coded names.
- Add focused checks for the argument-sensitive syntax strings:
  `sparse_grams` must mention `min_ngram_length`/`max_ngram_length`, and
  `vector_similarity` must include the required `dimensions` argument.
- Guard the `vector_similarity` check so it passes whether or not that
  type is registered in the current build (it requires USearch).

Also list `system.data_skipping_index_types` in
`02117_show_create_table_system` as suggested in review.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>