fix: expose MRTR state to tool handlers

[DaleSeo]

Fixes #1095.

Motivation and Context

ToolCallContext::new was dropping request_state and input_responses, so #[tool] handlers could not continue an MRTR call after the client completed its input requests. Because of that, servers had to bypass the macro routing path and implement ServerHandler::call_tool manually.

This PR preserves the SEP-2322 retry state and input responses when macro-generated tool handlers route calls through ToolCallContext. It also adds extractors for both values and lets #[tool] handlers return CallToolResponse. This allows a single handler to produce both input_required responses and completed results.

How Has This Been Tested?

Added regression tests

Breaking Changes

None.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update

Checklist

• I have read the MCP Documentation
• My code follows the repository's style guidelines
• New and existing tests pass locally
• I have added appropriate error handling
• I have added or updated documentation as needed

[alexhancock]

I'd strip these comments

[DaleSeo]

@alexhancock Would you be open to documenting the existing fields instead? Since this struct is public, I think rustdoc for all of its fields would be useful and more consistent.

[DaleSeo]

I've added comments to the existing fields.