Add batch vector insertion API to C ABI (FFI)

[tanvincible]

Description:

The current C ABI exposes only single-vector insertion:

uint64_t chassis_add(ChassisIndex*, const float* vector, size_t len);

For high-throughput ingestion and minimizing FFI overhead, we need a batch insertion interface.

FFI call overhead dominates when inserting vectors one-by-one from Python / Node / C++.

Proposed API:

size_t chassis_add_batch(
    ChassisIndex* index,
    const float* vectors,   // contiguous array
    size_t count,           // number of vectors
    size_t dim,             // dimension per vector
    uint64_t* out_ids       // output buffer of size >= count
);

Semantics:

• Inserts count vectors of dimension dim

• Vectors are laid out as:

  [v0_0, v0_1, ..., v0_dim-1,
   v1_0, v1_1, ..., v1_dim-1,
   ...]
  

• Returns number of vectors successfully inserted

• On partial failure:

  • Stops at first error
  • Sets last error message
  • Returns number inserted so far

Why This Matters:

• Enables efficient Python bindings
• Enables bulk ingestion pipelines
• Reduces FFI crossings by 10–100x
• Aligns with existing internal batch-friendly storage and graph logic

Implementation Sketch:

• Validate pointers and dimensions

• Loop over slices:

  for i in 0..count {
      let slice = &vectors[i*dim .. (i+1)*dim];
      index.add(slice)?;
  }

• Fill out_ids[i] with returned ID

Acceptance Criteria:

• New C ABI function implemented
• Header generated via cbindgen
• FFI tests covering batch insert
• No performance regression for single insert path