Sandbox host hygiene: deploy-time Docker prune and stale instance id in docs

Raw intake from the R6.3 on-call CI check (issue #21, run 34689637462).

• The sandbox deploy failed once with 'No space left on device' during 'uv sync --frozen' in the Docker build: the 8 GB root volume had accumulated 14 Docker images (1.25 GB reclaimable) and 51 MB build cache. On-call pruned via SSM and the rerun went green.
• Ask: add a prune step to the sandbox deploy flow (e.g. 'docker image prune -f' / builder-cache prune in or before scripts/deploy_relay_sandbox.sh), a scheduled cleanup, or a larger volume. Volume was ~73% after the prune and will fill again as releases accumulate.
• Docs drift: the instance id in docs/context.md (i-044716a7d6b053eef) differs from the instance the deploy workflow targets (i-03b7cd5de0a10a889). Update the doc or the workflow whichever is wrong.

Needs grooming.

Start by reading scripts/deploy_relay_sandbox.sh and the sandbox deploy workflow to confirm where Docker cleanup can safely occur. Then compare the instance IDs in docs/context.md and the workflow target. Done means the deploy flow has an agreed cleanup approach and the documentation or workflow uses the correct instance ID, with the relevant deploy check rerun.