Add Foundry IQ knowledge base support

[therealjohn]

Summary

Add support for Foundry IQ knowledge base scenarios in the Azure Developer CLI Foundry extension flow.

References

• Docs PR for Foundry IQ knowledge base hosted-agent quickstart: https://github.com/MicrosoftDocs/azure-ai-docs-pr/pull/12848
• Sample PR showing the Foundry IQ knowledge base agent implementation: https://github.com/microsoft-foundry/foundry-samples-pr/pull/570

Notes

Use the sample implementation as the reference for provisioning/configuring the knowledge base and wiring it into the hosted-agent experience.

[glharper]

Investigation: this scenario appears already supported on main

I traced the Foundry IQ knowledge base flow from both referenced PRs (docs MicrosoftDocs/azure-ai-docs-pr#12848 and sample microsoft-foundry/foundry-samples-pr#570) against the current extension code. The KB provisioning and wiring (the provision_kb.py script, toolbox.yaml, agent.manifest.yaml, and postprovision hooks) live in the foundry-samples repo by design; the azd-side commands they depend on already exist on main:

Step (hook + quickstart)	azd support
azd ai agent init -m <manifest-url>	✅ container-agent init copies/downloads the full sample dir (hooks, toolbox.yaml, provision_kb.py)
azd ai connection create --kind remote-tool --force --target … --auth-type agentic-identity --audience https://search.azure.com --metadata "ApiType=Azure"	✅ all flags present; agentic-identity routes through the raw-REST path carrying Audience/Metadata (azure.ai.connections/internal/cmd/connection.go)
azd ai toolbox create knowledge-base --from-file toolbox.yaml (an mcp tool with project_connection_id, server_url, server_label, require_approval)	✅ validateRawToolEntries passes the mcp entry through verbatim; toolEntryReferences explicitly recognizes mcp project_connection_id (azure.ai.toolboxes)
azd ai toolbox show knowledge-base	✅ present
azd ai agent run / invoke, azd provision / deploy / down	✅ standard flows

I also ran the sample's exact toolbox.yaml through the real parseToolboxFile → validateRawToolEntries → dup-check path, and it was accepted with every field preserved (type=mcp, project_connection_id=knowledge-base-mcp, require_approval=never). This matches the sample PR's own statement that it was validated end-to-end with these exact azd commands, and recent toolbox work (#8428 tools[] passthrough, #8677 env sync).

Conclusion: based on code inspection, no azd extension change is required for this scenario — it works with the existing azd ai connection create / toolbox create / toolbox show surface plus the sample's bundled script and hooks.

Could a maintainer confirm whether a specific azd-side enhancement was intended here (e.g., a first-party azd ai knowledge-base provisioning command, or auto-wiring the samples postprovisionhook duringazd ai agent init`)? Both would be net-new features that go beyond the referenced sample (which deliberately provisions the KB via a Python script and documents manual hook wiring), so I want to confirm scope before any implementation. Otherwise this can likely be closed as already-supported.

[hund030]

This scenario is already supported.
User follows the Readme can deploy and run an agent with Fonndry IQ knowledge base connection.