fix(config): show reflects the #3461 object-store project-config fallback

[worktrunk-bot]

Problem

wt config show didn't reflect the object-store project-config fallback that #3461 introduced. In the exact state that fix targets — a bare repo whose default branch is checked out in no worktree — config show reported no project config while wt switch resolved and ran the default branch's committed hooks. The command meant to be the user's window into effective config contradicted what runs.

Root cause: config show read project_config_path() and load_project_config() independently. load_project_config() already falls back to the committed default-branch config via git show <default>:.config/wt.toml, but project_config_path() still returns the (missing) on-disk path with no knowledge of that fallback.

• JSON emitted an internally inconsistent object: "exists": false alongside a populated "config": { … }, with "path" pointing at a file that isn't the real source.
• Text keyed off project_config_path() alone: it printed Not found and returned without ever showing the config that is actually loaded.

This is a diagnostics inconsistency only — hooks still run correctly. But #3461's motivating scenario is an agent-driven workflow that parks the primary off-branch, so it's exactly the users who'd inspect config show to understand hook behavior, and it told them the opposite of reality.

Solution

Both render paths now consult the object-store fallback when no on-disk file resolves:

• JSON: when config resolved but not from an existing on-disk file, path surfaces the revision spec (main:.config/wt.toml) and exists is keyed off the loaded config rather than path.exists(), so path/exists/config agree.
• Text: resolves the effective source the same way ProjectConfig::load does — on-disk file when present, else the committed default-branch config from the object store — labeled @ main:.config/wt.toml (from object store), and dumps the config through the existing deprecation/validation rendering.

Non-fallback cases are unchanged: exists still equals "config resolved" for on-disk configs (an empty file loads to a default Some, a parse error already errors the command), so no existing snapshot or JSON assertion shifts.

Testing

• New regression test test_bare_repo_config_show_reflects_object_store_fallback sets up the config-present parked state (bare repo, main carries .config/wt.toml, primary parked on feature-x, invoked from a linked worktree whose branch git rm'd the config) and asserts both JSON (path/exists/config agree) and text ((from object store) label, config dumped, never Not found). This case was previously uncovered.
• Confirmed the test fails on the unfixed code: {"config":{"post-start":"echo hi"},"exists":false,"path":".../other/.config/wt.toml"}.
• Full config_show (162) and bare_repository modules pass — no snapshot churn.
• cargo clippy --bins --lib clean.

Maintainer call on presentation

The original issue flagged the exact source label as a judgment call. This PR proposes @ main:.config/wt.toml (from object store) for text and the raw revision spec for JSON path; both are easy to adjust if you'd prefer different wording. The correctness change — showing the config that actually loads — is the substance; the label is cosmetic.

---

Closes #3476 — automated triage

[worktrunk-bot]

Self-review. The correctness change is sound — both render paths now consult the object-store fallback, the "No project config" vs Not found distinction survives the restructure, and exists = project_config.is_some() stays consistent with the on-disk cases (empty file → default Some, missing → None). One coverage nit worth pre-empting the codecov/patch gate: the .or_else(|| on_disk.clone()) fallback in the JSON path is unreachable and will show as an uncovered patch line — inline suggestion below.

[worktrunk-bot]

This arm is entered only when config.is_some() and no on-disk file resolved — which in this codebase means the config was loaded from the object store (ProjectConfig::load returns Some only via on-disk or default_branch_project_config_content). So within a single invocation default_branch_project_config_content() here is guaranteed to return Some, and .or_else(|| on_disk.clone()) never runs. It's dead defensively, and since the new test (like every reachable state) always takes the Some path, the closure body will show as an uncovered patch line on codecov/patch. Dropping it also removes the only way path/exists could disagree, which is the invariant the PR is establishing.

Suggested change

	_ if config.is_some() => repo
	.default_branch_project_config_content()
	.map(|(_, spec)| spec)
	.or_else(|| on_disk.clone()),
	_ if config.is_some() => repo
	.default_branch_project_config_content()
	.map(|(_, spec)| spec),

[worktrunk-bot]

Applied in 74f851f. Dropped .or_else(|| on_disk.clone()) — as noted, the arm is entered only when config.is_some() with no on-disk file, i.e. config came from the object store, so default_branch_project_config_content() is guaranteed Some and the fallback was dead. Removing it also removes the only way path/exists could disagree, which is the invariant this PR establishes.

config_show module still green (134 passed).