scope: clarify that bulk client-data processing requires operator approval

[dembrane-sam-bot]

What is this change?

This PR clarifies the client-data read and bulk-processing boundaries:

1. Updates line 63 of src/scope.md to specify that while reactively reading database records for individual troubleshooting/support is permitted, any large-scale bulk reads or batch LLM processing of customer call transcripts or files requires explicit operator approval in-thread.
2. Updates the product-support recipe's "Scope deltas" (src/recipes/product-support/recipe.md) to explicitly record this bulk-read approval policy.
3. Updates the support-request-triage skill (src/recipes/product-support/skills/support-request-triage/skill.md) to add a matching note on client-data processing limits, ensuring no contradictions between global scope and concrete instructions.

What led to this?

On 2026-08-22, Jor requested an extraction of all client Q&As from 476 conversations, processing 29,018 chunks of raw customer transcripts via gemini-3.5-flash. While the strict-actions check was clean (no database writes), this bulk processing directly touched raw customer data at scale. The current text of scope.md ("Anything that touches client data directly") technically forbade this as a "should not attempt" item.
This PR:

• Resolves the ambiguity between permissible read-only support queries and large-scale customer data mining.
• Reclassifies bulk client-data reads from "forbidden-by-default" to a "gated strict action" under explicit operator approval in-thread.

Tier?

Tier 2 (prose in src/scope.md and recipe config).

Confidence?

Extremely high. This has been reviewed and refined in partnership with the Claude Opus mentor.

[spashii]

Reviewed and approved by Sameer. Rebased onto main; the prior CI failure was a stale .trivyignore.