Regenerate AgentServer models from Foundry release - #48794

Summary

• advance the AgentServer TypeSpec pin to the consolidated Foundry release merge
• regenerate the complete TypedDict model surface and request validators using the SDK-pinned TypeSpec dependencies
• add memory-command, procedural-memory, model-routing, content-filter, expanded tool configuration, and shell streaming-event contracts
• refresh the API review surface and changelog for 2.2.0b2

Why the generated surface changed

The previous SDK pin (be2980b2e3b8fbf9e2034cbe441c388b2f6b54b2) predates the Foundry release consolidation. The new pin (9a69dbbfe1adfc1dc7f2a0ef4fa374dadf8abf7b) combines feature/foundry-release with the AgentServer projection, so generation now consumes the current shared OpenAI, memory, tool, toolbox, agent, skill, and common model sources.

Generation continues to use the SDK-pinned @azure-tools/openai-typespec 1.25.0 and @azure-tools/typespec-python 0.63.5. The generated differences therefore come from the updated Foundry source contracts rather than an emitter dependency update.

Contract compatibility

• output-text annotations and logprobs remain optional
• function-call output call_id remains optional
• ResponseUsageInputTokensDetails.cache_write_tokens remains required

Validation

• full package tests: 1,529 passed, 6 skipped
• TypeSpec model and OpenAPI generation completed successfully
• request validators regenerated from the emitted OpenAPI document
• API markdown and metadata regenerated successfully

Spec source: Azure/azure-rest-api-specs@9a69dbb