feat(targets): add native target for Azure AI Foundry v2 Agents (UUID IDs)

[tangym]

Problem you're trying to solve

PR #252 added native support for the v1 surface (asst_xxxx IDs) via target.model: azure_ai/agents/<asst_id>, routed through LiteLLM's azure_ai/agents provider with AAD/SP/MI auth.

The newer v2 Foundry Agents API (UUID IDs, served at {project}/agents/{uuid}/threads/...) is not supported by that path. Current workarounds are

• target.callable wrapper in PR Add LangGraph Foundry hosted-agent example with Adaptive Eval #250: user writes ~50 lines of glue per agent
• Wait for LiteLLM upstream support: out of our control, no ETA

Proposed solution or behavior

A small native target type for v2 hosted agents that mirrors the shape of target.endpoint: single config field, no callable wrapper, AAD via the same azure_auth helper this PR introduced. Tentative YAML shape (open to alternatives):

target:
  foundry_agent:
    project_endpoint: https://<resource>.services.ai.azure.com/api/projects/<project>
    agent_id: <uuid>

Runtime would build a session class similar to HTTPEndpointSession that speaks the v2 REST protocol (create thread, post message, create run, poll, read messages) using azure_auth.get_azure_token_provider(AZURE_FOUNDRY_SCOPE) for auth.

Alternatives considered

No response

Use case context

No response

Acknowledgements

• I searched existing issues and discussions and did not find a duplicate.
• I'd be willing to contribute a PR for this (optional).

[tangym]

Doc coverage parked for this issue

I noticed the doc surfaces left stale by #252 fall into two buckets, and the
second is the right home for them once v2 lands here:

Already in draft PR #256:

• assert_ai/internal_pipeline_prompts/init_system.md — parse-error
  prevention so assert-ai init doesn't emit target.tools or
  target.system_prompt alongside azure_ai/agents/<id> (the parser
  rejects both). v1/v2-agnostic — could ship today.
• docs/targets/README.md — adds Foundry-agent rows to the
  decision-tree and "paths at a glance" tables.
• docs/targets/model-and-tools.md — adds a "Foundry hosted agent
  target" subsection with a runnable YAML snippet and auth requirements.

The two docs/targets/* commits are held until this issue ships, since
promoting the route in user-facing target docs while v2 still needs a
workaround would oversell coverage.

Pickup path for this issue:

1. Rebase docs: post-#252 reconciliation for azure_ai/agents/<id> (init prompt + targets docs) [blocked on #253] #256 on top of the v2 implementation.
2. Tweak the docs to cover both v1 and v2 ID shapes (and any UX
   differences — UUID format, etc.).
3. Flip docs: post-#252 reconciliation for azure_ai/agents/<id> (init prompt + targets docs) [blocked on #253] #256 to ready for review.

If anything in the existing draft becomes outdated by the v2 design,
just drop or amend the commit — the prose is a starting point, not a
contract.

Cross-link: #256