Move the consolidated labs to the Caldova scenario

[ivorb]

Moves the consolidated labs (A–D) from Tailwind Traders to Caldova, the pharmaceutical scenario from the case study draft.

Caldova faces a 7% capacity gap across three manufacturing sites ahead of an accelerated launch, and needs to decide whether it can close the gap internally or engage a pre-qualified contract manufacturer.

The teaching mechanics don't change

Every function tool keeps its arity and its arithmetic, so the JSON tool schemas, the hand-written tool-calling loop, and the Agent Framework contrast all still demonstrate exactly what they did before:

Before	After
next_available_trip(region)	next_available_slot(site)
calculate_rental_cost(gear_tier, days, service_level)	calculate_transfer_cost(cmo_tier, weeks, priority)
generate_booking_report(...)	generate_capacity_report(...)
get_weekly_sales()	get_weekly_consumption()

Rates and multipliers keep their values, so the worked examples in the instructions stay arithmetically correct — 5 × 300 × 1.25 is still 1,875.

Actions draft rather than commit

The case study lists change control over production systems and supplier portals as a customer concern. Where the retail labs had the agent commit an action (take a booking), the Caldova equivalent drafts a request for a person to approve. No code shape changes, and it makes the lab more credible in a regulated setting.

Content rewritten, not renamed

• Six knowledge articles replaced with plant capacity, CMO directory, tech-transfer playbook, capacity request policy, supplier guide, and site operations.
• Lab D's evaluation set rewritten against those articles so every ground_truth is checkable against real source text.
• Red-team objectives stay in the same three SDK risk categories but move to a manufacturing setting. The self-harm objective is anchored in workplace distress rather than medication, which is the obvious but unacceptable pharma analogue.
• Lab C agents become tech-transfer planning, site-visit expenses, and supplier/quality triage.

Scope

• Instructions/Exercises is untouched — the published Learn labs never referenced the scenario.
• Anton is untouched. He's the Ask Anton learning guide, not a character in the scenario.
• Cora and Litware from the source doc are deliberately excluded.
• Personas appear as roles only; no first names in lab content.

Naming

Invented names (three sites, three CMOs, the CAL-204 program code) aren't supplied by the case study and may change. They're recorded together so a future rename is a single pass. All fictional domains use .example, which is reserved and can't resolve.

The five azd environments are renamed (tailwind-traders-lab → caldova-lab, and so on), so anyone mid-deployment gets a fresh environment.

Verification

All Tier 0 checks pass, plus generate_lab_blocks --check and sync --check. Every Python file parses, all JSON/JSONL/YAML validates, both function trees were executed and return correct values including error paths, and every file referenced by the instructions resolves. A vocabulary sweep for retail terms across instructions and Labfiles returns zero.

Not yet done: nothing here has been run against a live Foundry project.