Configure stop words for milli search index

[monneyboi]

Problem

The get_collection_terms tool returns common words like "the", "and", "from" which aren't useful for understanding collection topics:

- the (15967 docs)
- and (12878 docs)
- from (11044 docs)
- for (9652 docs)

Solution

Configure stop words in milli using settings.set_stop_words() in open_index().

Considerations

• Milli's stop word API takes a single global set, not per-language lists

• Options:

  1. Multilingual stop word set - Combine stop words from common languages (EN, ES, FR, DE, etc.)
  2. Per-collection language setting - Let users specify language(s) when creating a collection
  3. Auto-detect - Detect languages in documents and build combined list

• Requires re-indexing existing documents after configuration change

• Milli uses charabia for multi-language tokenization, but stop words are separate

References

• Milli stop words API: ~/.cargo/git/checkouts/meilisearch-77f25ebad2a1f1e5/aee74f4/crates/milli/src/update/settings.rs
• Could use stop-words crate for standard lists