Handle MCP blob resources without MIME types safely

[sylvesterkaczmarek]

Summary

Prevent MCP binary blob resources with no MIME type from being treated as UTF-8 text.

mcp_resource_to_content() currently considers mime_type=None supported for every MCP resource. For a BlobResourceContents, that causes arbitrary binary bytes to be base64-decoded and then decoded as UTF-8, which can raise UnicodeDecodeError. It can also cause an unknown binary blob to be selected ahead of a later supported resource in the same ReadResourceResult.

Fix

• treat missing MIME as text only for TextResourceContents;
• skip no-MIME blob resources while selecting the first representable resource;
• raise UnsupportedMCPValueError for a directly embedded no-MIME blob instead of leaking a Unicode decoding exception;
• preserve explicitly typed text/*, PDF, and supported image behavior;
• include <missing> in the unsupported-resource diagnostic when MIME metadata is absent.

Regression coverage

Adds offline tests covering:

• an isolated no-MIME binary blob;
• a no-MIME blob followed by a usable text resource;
• an embedded no-MIME binary resource;
• unchanged support for an explicitly text/plain blob.

The change is confined to the hand-maintained MCP conversion helper and its tests.