fix(send): show failure reason and IDs after response streaming

[adrianwebb]

Outcome

Human CLI output shows the actual failed target reason plus send and assignment IDs, rather than pointing to an invisible receipt. Fixes #191.

Work authority

• Work item / Issue: Show failed agent reason and correlation IDs in human send output #191; Capture pre-application Kata guest startup failures without restoring FIFO I/O deployment#695
• Proposal / decision: Preserve usable user-visible failure diagnostics.
• Assignment / checkpoint: User's repeated exact send failure.
• Actor: adrianwebb
• Human authority: User requested reliable execution and better debugging.
• Agent / capacity provider: Codex desktop under user authority.
• Exact base ref: f5b1d26
• Exact head ref: 5f3c6f6
• Agent-authored under human authority

Plan

Render failed targets even when responses were streamed, retain response deduplication, bound text and remove control characters, test the user's failure shape.

Changes and commits

5f3c6f6 adds compact failure rendering and a streamed receipt regression.

Verification

Three focused outcome tests pass; source architecture checks pass. Managed CLI rebuild running; Actions pending.

Risk and rollback

Display-only change; no credential or schema migration. Render bounded failure fields, not protected execution context. Revert through managed development if necessary.

Completion summary

The previously invisible human failure is implemented; underlying intermittent Kata startup is still tracked separately, not declared fixed.

Submission checklist

• The change is bounded to the stated work item and target repository.
• Exact base and head refs are recorded and the branch is ready for review.
• Verification and compatibility evidence are recorded above.
• No plaintext secrets, credentials, machine state, or unrelated residue are included.
• Plan, status, commits, and completion summary form a complete durable record.
• Rollback or recovery steps are documented and executable.