WOR-1226 add sidecar semantic cache e2e

[rickcrawford]

Summary

• add a local-only semantic-cache sidecar e2e fixture and integration test gated on SBPROXY_TEST_EMBED_MODEL plus SBPROXY_TEST_EMBED_TOKENIZER
• launch the release classifier sidecar with --embed-model, drive two near-duplicate chat prompts through ProxyHarness, and assert the second response is served from semantic cache without a second upstream chat call
• update the AI gateway semantic-cache docs to use action.semantic_cache, link local inference, and regenerate docs/llms-full.txt

Validation

• cargo test -p sbproxy-e2e --test semantic_cache_sidecar_e2e --locked
• cargo test -p sbproxy-config --test validate_examples --locked
• ./scripts/check-config-schema.sh
• ./scripts/regen-llms-full.sh --check
• cargo clippy -p sbproxy-e2e --test semantic_cache_sidecar_e2e --locked -- -D warnings
• git diff --check
• SBPROXY_CLEAN_AFTER_BUILD=0 ./scripts/check.sh

NOTICE

No new third-party libraries were added; NOTICE is unchanged.