Add prompt repetition usage guide (PR3)

[qicesun]

What

• Add prompt repetition guide in langchain4j-community-core/README.md
• Add root README link to this guide

Guide covers

• Non-RAG setup with PromptRepeatingInputGuardrail
• RAG setup with RepeatingQueryTransformer
• AUTO mode boundaries
• Idempotence behavior
• A/B rollout checklist

Validation

• ./mvnw -pl langchain4j-community-core verify -DskipITs

Depends on #571 and #572
Related #570

[qicesun]

Validation results (using this PR code, gpt-4.1-mini, Feb 16, 2026)

Functional

• langchain4j-community-core tests: 108/108 passed
• End-to-end checks passed:
  • rewrite works in ALWAYS
  • AUTO skips reasoning-intent prompts
  • idempotence works
  • input guardrail skips when RAG augmentation exists
  • query transformer preserves metadata
  • query transformer runs in retrieval augmentor flow

A/B: Non-RAG (mode=ALWAYS)

Bucket	Cases	Exact A	Exact B	Delta
NameIndex	120	61.67%	79.17%	+17.50pp
OptionsFirst	120	80.00%	100.00%	+20.00pp
MiddleMatch	120	97.50%	100.00%	+2.50pp
Aggregate	360	79.72%	93.06%	+13.33pp

A/B: Reasoning control (mode=AUTO)

• Cases: 80
• Applied rate: 0%
• Reasons: SKIPPED_REASONING_INTENT=80

A/B: RAG (mode=AUTO, query-side repetition)

• Cases: 30
• Exact: 1.00 -> 1.00

Conclusion: feature behavior is correct, and non-RAG objective tasks improve.

[qicesun]

Quick update: I opened #592 as the new draft PR that carries prompt repetition forward in the dedicated langchain4j-community-prompt-repetition module requested during review.

I also found an upstream langchain4j bug in AiServices input-guardrail sequencing:

• bug: [BUG] AiServices input guardrails run before content arguments are materialized into the final UserMessage langchain4j#4678
• draft fix PR: fix: materialize content arguments before input guardrails langchain4j#4679

Because of that, #592 is now the main review target and this PR is effectively superseded by the dedicated-module draft plus the upstream fix.