fix(llm): preserve semantic embedding batch order

[JesusMan0529]

When more than 1,000 new vertices are indexed, embedding batches can finish out of order. BuildSemanticIndex currently flattens those results in completion order but pairs them with the original vertex IDs, so semantic search can return the wrong vertex.

Collect batch results with asyncio.gather to retain input order. Keep the existing batch size, concurrency limit, synchronous embedding provider calls, and progress updates as each batch completes.

The regression forces the trailing batch to finish before the first batch and checks real Faiss save/load/search for both PRIMARY_KEY and CUSTOMIZE IDs. It fails on the original code. Empty input and provider error propagation are also covered.

Validation (Python 3.11):

• SKIP_EXTERNAL_SERVICES=true uv run pytest hugegraph-llm/src/tests/operators/index_op/ hugegraph-llm/src/tests/indices/ -q --tb=short: 53 passed, 1 skipped (the existing live Ollama test).
• uv run ruff format --check .: passed.
• uv run ruff check .: passed.
• git diff --check: passed.

HugeGraph Server and live LLM/provider integration tests were not run locally.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[JesusMan0529]

The regression uses 1,001 vertices and forces the one-vertex trailing batch to complete before the first 1,000-vertex batch. It checks nearest-neighbor lookup after saving and reloading a real Faiss index, for both PRIMARY_KEY and CUSTOMIZE ID strategies; it fails on the original operator and passes with this change. Batch completion still advances the progress bar immediately, and empty input and provider errors are covered. The related index-operator/Faiss suite passed with 53 tests and one existing Ollama-service test skipped.

[JesusMan0529]

Hope it could be merged 😊

[imbajin]

thx