Feat/build index from arrays

[sjswerdloff]

Adds a new method that accepts numpy arrays directly, avoiding the pickle intermediary. Refactors build_index_from_embeddings to delegate to it. Useful for MLX, GPU, or database-sourced embeddings.
Enables incremental embedding, prior to final indexing, e.g. for input, thinking, output turns in an LLM conversation, or adding the embeddings of one compaction/compression cycle of a long lived/many cycle conversation rather than recalculating all of the embeddings for each cycle. Reduces risk of failed modification (read/write/insert/write) to a pickle file when using an ACID database that has a vector data type.

Includes 7 tests covering basic usage, regression, dimension/count validation, and placeholder generation.

What does this PR do?

Adds a new method that accepts numpy arrays directly, avoiding the pickle intermediary. Refactors build_index_from_embeddings to delegate to it.

Related Issues

Fixes #

Checklist

• Tests pass (uv run pytest)
• Code formatted (ruff format and ruff check)
• Pre-commit hooks pass (pre-commit run --all-files)

[sjswerdloff]

The build error doesn't have anything to do with python (and the code change was purely python).
I'll look to see if previous PRs are having similar problems...
Any suggestions much appreciated.

CMake Error at /tmp/tmpmh94a5_1/build/_deps/protobuf-src/cmake/libprotobuf-lite.cmake:27 (target_link_libraries):
Target "libprotobuf-lite" links to:

absl::absl_check

but the target was not found. Possible reasons include:

* There is a typo in the target name.
* A find_package call is missing for an IMPORTED target.
* An ALIAS target is missing.

Call Stack (most recent call first):
/tmp/tmpmh94a5_1/build/_deps/protobuf-src/CMakeLists.txt:278 (include)

[ASuresh0524]

Thanks for the contribution. The build_index_from_arrays changes look good, and this failure appears unrelated to your Python changes.

The failing job is in packages/leann-backend-diskann CMake config (tcmalloc / absl::* target resolution), which is an environment/dependency issue in the DiskANN native build path, not in api.py / test logic from this PR.

I’ve seen the same failure pattern on other PRs:

• absl::absl_check / absl::nullability not found
• failure inside tcmalloc/protobuf CMake generation
• matrix cancellations after the first platform failure

I’m handling this as a separate backend/CI fix. Once that lands (or after rebasing onto a commit that includes it), please re-run checks on this PR.

In short: no blocker from the feature itself; current red CI is an infra/backend build issue.

[sjswerdloff]

There was a rebuild attempt. Same or very similar error:
CMake Error at /tmp/tmpyhuyq3g3/build/_deps/protobuf-src/cmake/libprotobuf-lite.cmake:27 (target_link_libraries):
Target "libprotobuf-lite" links to:

absl::absl_check

[sjswerdloff]

All checks passed!
Please merge at your earliest convenience.