What this PR does / why we need it:

Adds two field tags so the Milvus online store can set up text analysis on a String feature when
it creates a collection, in the same way milvus.index declares scalar indexes (#7120):

Field(
    name="title",
    dtype=String,
    tags={
        "milvus.analyzer_params": '{"tokenizer": "whitespace"}',
        "milvus.enable_match": "true",
    },
),

• milvus.analyzer_params: JSON object with the field's analyzer. Turns on enable_analyzer.
• milvus.enable_match: "true" builds the token index that Milvus TEXT_MATCH filters need.
• Both apply only to String features and need Milvus 2.5 or later.
• Feast does not query the tokens: Feast filters have no TEXT_MATCH operator and
  retrieve_online_documents_v2 still filters a query_string with LIKE. Clients that query
  the collection directly can use them. A follow-up will add BM25 field tags on top of the
  analyzer.
• Like the other creation-time tags, the settings apply only when Feast creates a collection.
  Unlike them, Feast does not compare them with an existing collection or warn when they differ.
• _nullable_tag becomes _boolean_tag(field, tag) so both boolean tags share one parser and
  error message.

Which issue(s) this PR fixes:

N/A

Checks

• I've made sure the tests are passing.
• My commits are signed off (git commit -s)
• My PR title follows conventional commits format

Testing Strategy

• Unit tests

• Integration tests

• Manual tests

• Testing is not required for this change

• Unit: the created schema per tag combination, and validation (non-String feature, invalid
  boolean, analyzer JSON that is not an object).

• Integration (test_milvus_remote.py, against Milvus 2.5.27, 2.6.27 and 3.0.2): a whitespace analyzer
  keeps wool-blend as one token, so TEXT_MATCH finds wool-blend and not wool.

• Milvus Lite accepts both tags but does not apply them (its TEXT_MATCH always uses the standard
  analyzer), so only the server tests cover the behaviour.

Misc

Milvus 2.6 makes new rows visible to TEXT_MATCH about 2 seconds after the write, even with
Strong consistency, so the integration test retries that query.

Written with @jonasfagerlund, who is a co-author of the commit. A BM25 PR builds on this one.