FIX Infer Azure Blob MIME types from file extensions

[romanlutz]

Description

Azure Blob payloads written through AzureBlobStorageIO currently receive the configured default text/plain metadata even when their filenames identify images, audio, video, or documents. This can cause downstream consumers to handle otherwise-correct bytes as text.

This change adds a private, deterministic extension-to-MIME mapping inside AzureBlobStorageIO.write_file_async, retaining the configured default for unknown or extensionless paths. It deliberately preserves the original two-argument StorageIO.write_file_async contract so existing third-party storage implementations remain compatible, and leaves AzureBlobStorageTarget's separate text-only content-type API unchanged.

Serializer filename handling now replaces an existing suffix instead of appending a duplicate one. Formatted PCM audio is always written as RIFF/WAVE with a .wav filename, allowing Azure metadata inference to set audio/wav, including when callers provide a filename with another extension.

This is a fresh replacement for the useful intent of closed PR #2233 without reusing its branch or broadening the public storage API.

Tests and Documentation

• Added real-behavior unit coverage for representative text, image, audio, video, and document MIME metadata.
• Added unknown-extension and extensionless fallback coverage.
• Added WAV filename, RIFF/WAVE content, and audio/wav metadata coverage.
• Added duplicate-extension and legacy custom StorageIO compatibility regressions.
• Ran the targeted storage unit tests: 86 passed.
• Ran targeted Ruff check and format validation; commit hooks including type checking also passed.
• Documentation changes: N/A; behavior is internal and the public API is unchanged.
• JupyText: N/A; no notebooks were changed.