…gic and improve resource disposal (apache#3649)

## Summary
This PR consolidates LZ4 decompression code paths and ensures proper
resource cleanup across both CloudFetch and non-CloudFetch readers in
the Databricks C# driver.

## Changes
- **Lz4Utilities.cs**
  - Add configurable buffer size parameter to `DecompressLz4()`
  - Add async variant `DecompressLz4Async()` for CloudFetch pipeline
  - Add proper `using` statements for MemoryStream disposal
  - Add default buffer size constant (80KB)

- **CloudFetchDownloader.cs**
  - Update to use shared `Lz4Utilities.DecompressLz4Async()`
  - Reduce code duplication (~12 lines consolidated)
  - Improve telemetry with compression ratio calculation

## Benefits
- **Code Quality**: Both code paths now share the same decompression
implementation, reducing duplication
- **Resource Management**: Explicit MemoryStream disposal improves
memory hygiene (though GC would handle cleanup)
- **Maintainability**: Single source of truth for LZ4 decompression
logic
- **Consistency**: Same error handling and telemetry patterns across
both paths

## Technical Details
- Default buffer size remains 80KB (81920 bytes) - no behavioral changes
- Async version returns `(byte[] buffer, int length)` tuple for
efficient MemoryStream wrapping in CloudFetch
- Buffer validity preserved after MemoryStream disposal via reference
counting
- Maintains cancellation token support in async path
- No performance impact - purely refactoring and cleanup

## Testing
- Existing unit tests pass
- No functional changes to decompression logic
- Telemetry output remains consistent

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude <[email protected]>