fix: recover from invalid default workspace paths

[Jordan-SkyLF]

Summary

Fix default workspace fallback so sessions do not retain a broken workspace path when the configured default is invalid or unavailable.

What this changes

• harden default workspace resolution in api/config.py
• add regression coverage for fallback behavior in tests/test_default_workspace_fallback.py

Why

When the default workspace cannot be used, the Web UI should recover to a safe fallback instead of carrying forward an unusable path.

Test plan

• python -m pytest tests/test_default_workspace_fallback.py -q
  • 2 passed

[nesquena-hermes]

PR Review: fix: recover from invalid default workspace paths

Thanks for this fix!

What this does

• Hardens the default workspace resolution logic in api/config.py so that when the configured default workspace path is invalid or unavailable, the Web UI falls back to a safe alternative
• Adds regression coverage in tests/test_default_workspace_fallback.py (2 passing tests)

Why this matters

Without this fix, users who have a stale or broken default workspace path would end up in a broken state where sessions inherit the unusable path. This is particularly impactful for users who have moved, deleted, or renamed their workspace directories.

Code quality notes

• The fix is minimal and well-targeted — hardening api/config.py is exactly the right place for this
• The test file name test_default_workspace_fallback.py is descriptive and the 2 tests confirm both the fallback trigger and the recovery behavior
• The cherry-pick commit history is clean

Summary

✅ Looks good to merge. The fallback behavior prevents a class of silent breakage where sessions silently carry forward an unusable workspace path.

[nesquena-hermes]

Merged into v0.50.18 ✅

Thanks @Jordan-SkyLF — this is a clean, well-scoped fix for a real pain point. The three-function refactor (_workspace_candidates, _ensure_workspace_dir, resolve_default_workspace) is easy to follow and the fallback priority order is sensible.

What was added before merging:

• Rebased onto current master (v0.50.17)
• 5 additional tests: ~/workspace creation when missing, RuntimeError when all candidates fail, dedup check on candidate list, env var priority, unwritable path returning False
• CHANGELOG entry

Now live in v0.50.18. Docker users and anyone with a stale workspace path in their settings will self-heal on next server start. 🎉

[nesquena]

Thanks @Jordan-SkyLF for the contribution!

After pulling the latest master and comparing line-by-line, every change in this PR is already merged:

• _workspace_candidates() — identical on master
• _ensure_workspace_dir() — identical on master
• resolve_default_workspace() — identical on master
• _discover_default_workspace() updated docstring — identical on master
• save_settings() workspace rewrite logic — identical on master
• Startup settings workspace fallback + settings.json rewrite — identical on master
• test_default_workspace_fallback.py — identical tests on master (plus additional tests)

This was merged as part of a prior sprint that addressed the same issue (#357). Closing as superseded — thanks for the effort!