Motivation & Context

CosmosChatHistoryProvider exposes utilities for counting and clearing stored messages, but applications cannot retrieve those messages without running the agent invocation pipeline. This makes migration, inspection, and other out-of-band conversation-history scenarios unnecessarily difficult.

Description & Review Guide

• What are the major changes?
  • Add GetMessagesAsync(AgentSession?, CancellationToken) to retrieve the messages stored for a session.
  • Reuse that method from ProvideChatHistoryAsync so public and invocation retrieval share the same Cosmos query, pagination, MaxItemCount, and MaxMessagesToRetrieve behavior.
  • Document that direct retrieval returns raw stored messages without the invocation pipeline's output filter or chat-history source attribution.
  • Dispose the Cosmos feed iterator and add coverage for pagination, empty histories, filtering/source attribution, disposal, and cancellation.
• What is the impact of these changes?
  • Consumers can inspect Cosmos-backed chat history without initiating an agent run.
  • Existing invocation behavior remains unchanged.
• What do you want reviewers to focus on?
  • The session-aware API shape and the documented distinction between raw retrieval and invocation-processed messages.

Related Issue

Fixes #3809

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
  • The Cosmos emulator-dependent tests are added and compile successfully, but were skipped locally because the emulator is unavailable. They should run in the emulator-enabled CI environment.
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue (see Related Issue above).
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.