feat(cortex): add trust visibility to cortex agents (#275 slice A)

[hurttlocker]

What this does

Implements a bounded #275 slice A for trust-boundary groundwork by making trust config visible in read-only operator surfaces.

• Adds trust config parsing + validation for agents.<id>.trust in ~/.cortex/config.yaml
• Surfaces trust_level + trust_scope in cortex agents --json
• Extends cortex agents table output with TRUST/SCOPE columns
• Explicitly labels this as read-only visibility (no write enforcement in this slice)

Problem / Context

Issue #275 needs scoped multi-agent trust boundaries, but the full system is too large for one PR. We needed the smallest safe first slice that:

1. validates trust config early, and
2. gives operators a deterministic read-only view of trust metadata.

How it works

Files touched

• internal/config/resolver.go
  • Added AgentTrustRule + AgentTrustEntry
  • Added ResolveAgentTrustConfig(configPath)
  • Added AgentTrustScope(trustLevel) with allowed values:
    • owner -> read:all write:all
    • collaborator -> read:all write:own
    • reader -> read:all write:none
  • Extended config schema with optional agents: map
• cmd/cortex/main.go
  • runAgents now loads/validates trust config before rendering output
  • agentSummary now includes trust visibility fields
  • TTY table now includes TRUST + SCOPE columns
  • Added explicit note that this is visibility-only (no cross-agent write enforcement yet)
• internal/config/resolver_test.go
  • Added deterministic tests for valid/invalid trust config parsing
• cmd/cortex/main_test.go
  • Added deterministic tests for JSON trust visibility, TTY trust visibility, and invalid trust config handling

Testing done

go test ./internal/config ./cmd/cortex
go test ./...

Screenshots / before-after

CLI surface change only (no UI). Example cortex agents output now includes TRUST/SCOPE columns and a read-only visibility note.

Breaking changes / risks

None.

Primary risk: cortex agents now fails fast when agents.<id>.trust is invalid in config. This is intentional validation behavior, but operators with malformed trust config will need to fix it.

Merge notes

• Bounded Multi-agent scoped views + trust boundaries #275 slice A only.
• Intentionally out of scope: cross-agent write enforcement, reconciliation engine, shared-fact promotion logic.
• Ready for Hawk deterministic gate before Q review.