fix(05,12): raise o3-deep-research TPM capacity from 10 to 200

[corticalstack]

Summary

Raise the o3-deep-research model deployment SKU capacity from 10 (10K TPM) to 200 (200K TPM) in both Bicep files that define it:

• 05-foundry-project-pattern-setup/05-02-deploy-foundry-core-gateway/main.bicep (the hub deploy)
• 12-foundry-iq-deep-research/main.bicep (the optional standalone deploy)

The original 10K cap throttles multi-step deep-research runs with 429 errors before they complete. 200K gives realistic headroom for the agentic deep-research loop while staying well under the Norway East o3-DeepResearch subscription quota (limit 3000 - confirmed via az cognitiveservices usage list -l norwayeast).

GlobalStandard billing is pay-per-token, so raising the capacity ceiling does not change baseline cost; it only raises the rate-limit cap.

Existing deployments

Bumping the Bicep value does not update already-deployed resources. To update a live deployment in place without redeploying the whole template:

az cognitiveservices account deployment update \
  -g rg-foundry-core-{suffix} \
  -n aif-research-{suffix} \
  --deployment-name o3-deep-research \
  --sku-capacity 200

Patch release 0.8.5.

Stacking note

This PR is stacked on fix/deep-research-apim-key-lookup (PR #12). Base set to that branch so the diff shows only the Bicep + version-bump changes. Merge #12 first, then this PR.

Test plan

• Fresh az deployment group create of either Bicep produces an o3-deep-research deployment with sku.capacity = 200
• 12-02-deep-research-loop.ipynb completes a multi-step research run without 429s
• az cognitiveservices usage list -l norwayeast still shows the OpenAI.GlobalStandard.o3-DeepResearch total well under its limit