Context-Mode Telemetry — Measuring Right-Sized Context

Broader goal (previous)

Platform ops could score hygiene vs real multi-agent work (windowed sentinels, job reap). The scoreboard still could not answer a product-critical question: do jobs get task-level or full context?

Why this PR

Right-sizing context by job type is a core synlynk claim. Without telemetry we only had defaults (--context-mode task) and folklore (PM often forces full). After a week of real by_context_mode data we can refine modes and, later, replace static generate_context with an Architect Workspace Agent as a configurable context provider.

What shipped

1. daemon_jobs.context_mode + context_bytes — written on every dispatch (after truncation).
2. jobs.json + telemetry — same fields on the job object / dispatch event.
3. cost_entries.context_mode — inherited from the job row on cost insert when not passed.
4. ops report — L1 by_context_mode counts + %, context_bytes summary; L2 cost by mode.
5. synlynk jobs table — CTX column.
6. Design note — docs/superpowers/specs/2026-08-09-context-mode-telemetry-and-architect-context-provider-design.md for the Architect provider path.

On the long arc

Instrumentation first → policy second → Architect-built context third. This PR is the measurement spine.

New goalpost

• Nightly ops shows mode mix (even if early data is mostly unknown + new task/full).
• After ~7 days: open Architect provider design implementation; mode policy informed by data.