fix(python): resolve multi-index search core dumps by exposing missing state (#643)

[Abhinav2656]

Description

Fixes #643: Mult-Index Search Doesn't Work in Python

Root Cause Analysis

When utilizing the Python bindings for multi-index search, ix.search(query) was failing violently via either a Python ValueError (dimension mismatch) or a C++ std::bad_function_call (core dump).

This occurred because the Indexes wrapper class dropped the state of its underlying Index children upon initialization:

1. Missing Metadata: The Indexes class lacked the ndim, dtype, and metric properties required by _search_in_compiled.
2. Dangling Pointers / GC: The Indexes.__init__ method consumed an iterable of Index objects, passed their compiled pointers to C++, and then immediately discarded the Python objects. If instantiated with anonymous objects, Python's garbage collector destroyed the underlying indices, leaving the C++ core with null pointers to metric functions (triggering the std::bad_function_call).

The Fix

• Bound the iterable of indices to self.indices within the Indexes constructor to ensure the Python objects (and their C++ metric callbacks) survive garbage collection.
• Implemented @property methods for ndim, dtype, and metric on the Indexes class that dynamically inherit the architectural state from the first underlying index in the collection.

Validation

Added test_multi_index to the Python test suite to strictly verify:

• Boundary Conditions: Empty Indexes([]) gracefully returns 0 / None / unknown.
• Recall Verification: Multi-index router successfully locates distinct vectors across different C++ shards.
• Batch Memory Alignment: Wrapper successfully handles batch matrix queries (np.vstack) across the boundary without dimension mismatch errors.