fix: rework search API generics to replace unnecessary type params with String

[viktoriya-kutsarova]

Summary

Reworks the generics across the RediSearch API to remove unnecessary type parameter exposure and replace them with String where values were never codec-encoded in the first place. Fixes incorrect codec usage in warning decoding across both search parsers.

Changes

Arguments classes — K and V removed

The following argument classes were fully de-genericized. All fields and builder methods that previously carried K or V become String, since they represent schema attribute names, filter expressions, stop words, prefixes, and other index-definition strings that are always UTF-8 text:

• CreateArgs<K, V> → CreateArgs (prefixes, filter, language field, score field, payload field, stop words)
• FieldArgs<K> → FieldArgs (field name, alias); all concrete subclasses follow: GeoFieldArgs, GeoshapeFieldArgs, NumericFieldArgs, TagFieldArgs, TextFieldArgs, VectorFieldArgs
• ExplainArgs<K, V> → ExplainArgs
• SpellCheckArgs<K, V> → SpellCheckArgs
• SynUpdateArgs<K, V> → SynUpdateArgs
• SugAddArgs<K, V> → SugAddArgs
• SugGetArgs<K, V> → SugGetArgs
• HybridArgs<K, V> → HybridArgs
• HighlightArgs<K, V> → HighlightArgs<K> — V removed (tags become String); K is retained because highlight fields are hash field identifiers, consistent with hget(K key, K field). Tags<V> → Tags, addValue() → add() in build()

AggregateArgs<K, V> → AggregateArgs<K>

V is removed entirely. Inside AggregateArgs, K is narrowed to LoadField only, since hash field identifiers are legitimately typed as K. As a consequence:

• LoadField.field is now correctly encoded via args.addKey() instead of the previous .toString() bypass; LoadField.alias becomes String
• All pipeline labels (Apply.name, Reducer.alias, GroupBy.properties, SortProperty.property, param keys) become String
• GroupBy, SortBy, SortProperty, Apply, Reducer, Filter, Limit and PipelineOperation are now non-generic with build() signatures narrowed from CommandArgs<K, ?> to CommandArgs<?, ?>

SearchReply and HybridReply warnings

warnings: List<V> → List<String> in both classes. Server-generated warning messages are always UTF-8 text and must not be decoded through the user-configured value codec. SearchReplyParser and HybridReplyParser are updated to use StringCodec.UTF8.decodeValue() at all warning-decoding sites — including both the RESP2 and RESP3 paths in HybridReplyParser, which had the same bug in two separate places.

Command interfaces and implementations

All method signatures updated across the full command surface:

• Sync: RediSearchCommands
• Async: RediSearchAsyncCommands
• Reactive: RediSearchReactiveCommands
• Cluster variants of all three
• AbstractRedisAsyncCommands, AbstractRedisReactiveCommands, RediSearchCommandBuilder
• Kotlin coroutine wrappers

Javadoc corrections

• Class-level examples in AggregateArgs corrected: wrong return type (SearchReply → AggregationReply), stale generic witnesses, invalid load() varargs call, and .reduce() chained directly on Builder instead of GroupBy
• WithCursor factory method Javadocs corrected: were copy-pasted from Apply with wrong @param descriptions and @return text
• filter() builder example corrected for the same .reduce() on Builder mistake

Testing

• All existing tests updated to reflect the new signatures
• Integration test testSearchWarningsReturnedAsStrings added to RediSearchIntegrationTests, verifying that getWarnings() returns List<String>, that the list is non-null for normal searches, and that any warning content produced under a 1 ms timeout is correctly decoded as a non-empty UTF-8 string
• Existing weak assertion on HybridReply.getWarnings() (size() >= 0) replaced with a typed List<String> assignment and per-element content validation

---

Note

Medium Risk
Public RediSearch command signatures are changed broadly (sync/async/reactive/cluster/builders), which is a source-compatible breaking change for clients and could affect downstream compilation and behavior. Runtime risk is moderate and mainly limited to parsing/encoding of RediSearch protocol literals (queries, warnings, dict/synonyms).

Overview
This PR reworks the RediSearch Java API types to stop treating query/dictionary/synonym/index metadata as user V values, replacing those parameters/return types with String across command interfaces and implementations (sync/async/reactive and cluster), and updating RediSearchCommandBuilder accordingly (e.g., ftSearch/ftAggregate/ftExplain/ftSpellcheck, dict/synonym ops, ftList).

It also fixes warning decoding in SearchReplyParser/HybridReplyParser and changes SearchReply/HybridReply warnings to List<String> so server warnings are decoded via StringCodec.UTF8 instead of the user value codec; SpellCheckResult is de-genericized to String-based terms/suggestions. Minor repo hygiene: .gitignore now ignores .augment and correctly matches .vscode*.

^{Reviewed by Cursor Bugbot for commit bb65286. Bugbot is set up for automated code reviews on this repo. Configure here.}

[tishun]

LGTM

(best effort since it as a huge review)

[tishun]

@a-TODO-rov TBH I am wondering if we should not just remove this. Perhaps @uglide can help us understand if there is a use case for keeping it?

[viktoriya-kutsarova]

Thank you @tishun for pointing this out. I will check with @uglide.

[viktoriya-kutsarova]

For some reason the AI agent has changed the groupBy notation from groupBy("category") to groupBy(GroupBy.of("category").

[viktoriya-kutsarova]

The AI agent has changed the example from load("price", "quantity", "category") to load("price").load("quantity").load("category").

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Reviewed by Cursor Bugbot for commit 0786cd8. Configure here.}