[Enhance]: Add output_fields support to fetch() for selective field retrieval

[chinaux]

Affected Component

Python/C/other API (and underlying C++ layer)

Current Behavior

The fetch() method currently retrieves all fields of a record when queried by primary key(s). Unlike query(), which supports the output_fields parameter to specify which fields to return, fetch() lacks this capability. As a result, users must download full records—including potentially large fields like document text or embedding vectors—even when only lightweight metadata (e.g., status, source_file, chunk_index) is needed. This leads to unnecessary network I/O, memory usage, and processing overhead.

Desired Improvement

Extend the fetch() method to accept an optional output_fields parameter, consistent with the existing query() interface. When provided, only the specified fields should be returned.

This requires:

• Adding the output_fields parameter to the Python/C API
• Propagating the field selection down to the storage layer (may require C++ changes)
• Ensuring backward compatibility: when output_fields is not specified, fetch() should continue to return all fields as before

Impact

• Reduced resource consumption: Avoid transferring and deserializing unused data (e.g., skip large embedding vectors or full document text when only metadata is needed).
• Improved performance: Faster response times for lightweight lookups by primary key.
• Better API consistency: Unified field-selection behavior between query() and fetch(), reducing cognitive load for developers.
• Enables efficient real-world workflows, such as:
  • Checking whether a file has been processed — only metadata fields are needed
  • Fetching neighboring chunks — only source_file and chunk_index are required
  • Tracking tunnel connections — only a small set of metadata fields are used for preview