Copy configured guest files before agent bootstrap

[ESultanik]

Add repeatable [[guest_files]] mappings to copy explicit host files or directories into each VM before agent bootstrap. Copies refresh on restart, retain guest-only files, and use private owner-writable permissions. --no-agents still copies the configured files.

Stage sources in private directories before VM startup or disk replacement. Allow source links only within declared roots, use descriptor-relative reads, and reject staging recursion, special files, destination overlaps or symlinks, and live-mount overlaps.

Update configuration examples and the configuration, command, architecture, trust, and testing references.

Stacked on #513.

Validation: workspace tests (1,200 coop, 30 proxy, and eight proxy integration), format, clippy, taplo, and normal pre-commit hooks passed. Mutation testing caught all 14 viable mutants (four unviable, none missed). Full Firecracker integration passed: 504 passed, zero failed, ten skipped.

The exact-candidate Lima run recorded 494 passed, one failed, and 13 skipped: an existing onboarding restart timed out waiting for SSH authentication. The unchanged isolated onboarding fixture then passed all eight checks. All six guest-file checks passed in the full attempt, and the complete combined stack at #516 subsequently passed full Lima integration (510 passed, zero failed, 13 skipped), including onboarding and all guest-file checks. The failed attempt remains recorded; it is not counted as a full pass.

[ESultanik]

Merge/readiness checklist for this PR:

Stack order: #512 → #513 → #514 → #516. Each PR should ultimately merge into main, in that order; do not merge a follow-on PR into its current feature-branch base.

After a predecessor merges, retarget the next PR to main (verify the base even if GitHub updates it automatically). Integrate main with a merge commit if needed, resolve conflicts, and check that the final diff contains only that PR's intended change. Preserve published history: no force-push/rebase is needed. Refresh required CI and review; repeat validation affected by any changes to the tested tree.

• Merge Run post-start hooks after workspace provisioning #513, then retarget this PR from fix/issue-511 to main.
• Complete Firecracker VM integration for this PR's candidate and record the results below.
• Confirm the final diff, required CI, and maintainer review; mark ready once platform gates pass, then merge into main.
• After merging, retarget Run resumable global creation hooks #516 to main.

Validation clarification: this PR's original full Lima run had one SSH failure during the existing stub-Claude image build; that phase passed on an isolated retry without code changes. All six guest-file assertions passed. The complete downstream stack at #516 commit 23cf11417bccfd4187c7b37ad0951ccc064d6f4b subsequently passed a clean full Lima run: 515 passed, zero failed, nine skipped, runner exit zero, including the marketplace phase. This is additional combined-stack evidence, not a replacement claim about this PR's original run. Workspace checks, scoped mutation testing, and Linux CI pass.

Use a clean checkout of this PR's final candidate and a Linux SSH host with working KVM/Firecracker access and the remote full-suite prerequisites. The full run needs the documented build/network tools and passwordless sudo. Run these suites serially on a shared host because they rebuild shared test images.

git rev-parse HEAD
./tests/run-integration.sh --full --remote user@host --name coop-pr-514

Replace user@host with the available test host. Confirm that guest-file copying, restart refresh, retained guest-only files, writable neighboring configuration, bootstrap ordering, and applicable mount protections execute and pass. Explain backend-specific skips.

Post the tested commit SHA, host architecture, runner exit status, pass/fail/skip counts, and a log link or attachment here. Inspect skipped phases; a zero exit alone is insufficient if relevant coverage was skipped. Linux CI does not substitute for Firecracker VM integration. A run of #516 alone provides combined-stack evidence, not proof that each earlier PR's exact candidate was tested.

[hbrodin]

Review summary for db0f4c8: two inline findings, covering reprovision test coverage and a same-size/same-mtime rsync refresh edge case. No diff-noise finding.

I reviewed correctness, design, conventions, security, API usage, tests, docs, and code comments. I checked the pinned diff and post-change call paths, ran git diff --check, syntax-checked the changed shell scripts, and reproduced the rsync behavior with the actual guest copy script. There were no earlier review threads to carry forward.

Current-head CI checks passed. I did not run Lima or Firecracker VM integration. The linked Firecracker full run exercised earlier commit 7ef0a67; the PR describes a Lima run with one failure and a later clean combined-stack run. Neither is a full VM run of this exact head.