docs: document read_blobs for blob payload reads

[Xuanwo]

Summary

This PR updates the blob documentation to make read_blobs the recommended path for workflows that need complete blob payloads in memory.

It also clarifies when to use take_blobs for lazy file-like access and when to use scanner(..., blob_handling="all_binary") for Arrow table scans.

Context

Blob users were likely to discover take_blobs first and then manually parallelize BlobFile.readall() calls for batch loaders. Lance already has read_blobs, which plans and executes batched materialized blob reads, but the guide did not surface it as the primary API for full-payload reads.

Validation

• git diff --check
• uv run make lint partially completed: Ruff format and Ruff lint passed.
• Pyright did not complete locally on macOS because the project installs tensorflow only through the Linux tests dependency marker; the remaining errors are existing unresolved tensorflow imports in python/lance/arrow.py, python/lance/dependencies.py, and python/tests/test_arrow.py.
• make install built the local pylance extension successfully, then failed at pre-commit install because this checkout inherits an existing core.hooksPath configuration.

[prrao87]

Fixed some styling nits to make it read better, but thanks, LGTM!