docs: post-#252 reconciliation for azure_ai/agents/<id> (init prompt + targets docs) [blocked on #253]

[tangym]

Summary

Reconciles the assert-ai init system prompt and the user-facing target docs with the azure_ai/agents/<AGENT_ID> Foundry hosted-agent route that #252 added.

Splits cleanly into two commits so the safe part can ship independently if needed:

1. docs(prompts): teach assert-ai init about azure_ai/agents/<id> — Parse-error prevention. The TargetConfig parser rejects target.tools and target.system_prompt when the model name starts with azure_ai/agents/, but the init system prompt did not learn this rule, so assert-ai init currently proposes configs that fail at parse time when a user describes a Foundry agent. v1/v2-agnostic — would be correct to ship today.

2. docs(targets): document Foundry hosted-agent target [BLOCKED on #253] — Promotes the Foundry route in docs/targets/README.md (decision-tree + "paths at a glance" tables) and adds a "Foundry hosted agent target" subsection to docs/targets/model-and-tools.md. Held until feat(targets): add native target for Azure AI Foundry v2 Agents (UUID IDs) #253 ships full Foundry v2 agent support. PR feat(azure-auth): native AAD support for azure_ai/* targets (Foundry hosted agents) #252 only covers v1, and the v2 path still needs a workaround, so promoting the route in public target docs now would oversell coverage.

Why draft?

Bidirectional pickup for #253:

• This PR → feat(targets): add native target for Azure AI Foundry v2 Agents (UUID IDs) #253: documents the dependency, so commit 2 doesn't ship prematurely.
• feat(targets): add native target for Azure AI Foundry v2 Agents (UUID IDs) #253 → this PR: gives whoever implements v2 a ready-to-rebase doc diff, so they don't have to re-derive what user-facing surfaces need updating.

When #253 lands, the v2 implementer should:

1. Rebase this branch.
2. Adjust commit 2's wording if needed (e.g. cover both v1 and v2 ID shapes).
3. Flip the PR to ready for review.

If we want commit 1 sooner, it can be cherry-picked into a tiny standalone ship-now PR without losing the draft.

Verification

• Re-read the updated decision tree end-to-end as if I were the init LLM, with two synthetic inputs:
  1. "I have a LangGraph agent in src/my_agent/bot.py" → routes to callable. ✓
  2. "I have a Foundry agent, agent ID asst_abc123" → routes to model sub-case 2b; does not ask for system_prompt/tools. ✓
• python3 -m compileall -q assert_ai (no code changed; sanity check).

Files

File	Commit
assert_ai/internal_pipeline_prompts/init_system.md	1
docs/targets/README.md	2 (blocked)
docs/targets/model-and-tools.md	2 (blocked)

Related: #252, #253