feat: make structured mirroring explicit

[DaleSeo]

Motivation and Context

Automatic mirroring can double large structured payloads and makes a compatibility fallback an implicit SDK policy. This implements the alternative discussed in #1046: structured output is the default, and callers that need text content choose it explicitly without adding another return-type wrapper or macro behavior.

How Has This Been Tested?

Added integration tests

Breaking changes

Technically none because the API hasn't changed.

However, code that relies on structured, structured_error, or Json<T> to automatically populate content now needs to add the fallback explicitly with with_content or construct the CallToolResult directly.

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]

cc @olaservo can you review?

[olaservo]

Hi @alexhancock @DaleSeo , thanks for putting together this alternative! After reviewing the surveyed clients again and considering that mirroring is a SHOULD in the spec, IMO it would be best to add an opt out for mirroring at this point, instead of an opt-in. And hopefully by version 4.0 we will figure out better protocol semantics for explicitly choosing the outputs and destinations. :)

For reference, here is a snapshot of what a sample of clients do with the different payloads: https://gist.github.com/olaservo/3bb819673c444e4fe282c3af44a1ae01

[DaleSeo]

Thanks, @olaservo, for the review and client survey. After comparing the current Tier 1 SDKs, I agree that keeping mirroring on by default with an explicit opt-out is the better direction for v3.
I'll close this in favor of your PR and help update it so we can move forward with the opt-out.