fix(e2e): build wt from the working tree instead of guessing the binary

[timvw]

Summary

• The e2e runner now builds wt from the working tree when --wt is not given, instead of scanning for a pre-existing bin/wt and then falling back to whatever wt is on PATH
• Both silent fallbacks are removed: they let a run report on a binary unrelated to the source being tested
• --wt is unchanged and remains the explicit escape hatch, which is what CI uses

Closes #141

Why

findWtBinary never built anything. It took the first of ./bin/wt, ./wt, bin/wt, then exec.LookPath("wt"). Both fallbacks misfired on a developer machine while verifying #139:

1. Stale bin/wt. bin/ is gitignored, so a build from earlier in the same worktree survives a rebase. After rebasing onto main (which had picked up fix(hooks): require approval before running hooks from a committed .wt.toml #134), the leftover binary had no trust command, and the two trust scenarios failed with unknown command "trust" for "wt".
2. PATH fallback. In a fresh worktree with no bin/, the runner silently selected /opt/homebrew/bin/wt — v0.1.33, predating both feat(clone): add wt clone with repo_root + repo_pattern placement #127 and fix(hooks): require approval before running hooks from a committed .wt.toml #134. Four scenarios failed there, since that build has no wt clone either.

In both cases the scenarios and the source were correct; the failures were an artefact of testing the wrong binary. The dangerous case is the inverse: a stale binary that predates a regression passes scenarios that should fail.

CI never hit this because the workflow always passes --wt=bin/wt, so the auto-detection path was exercised only by developers running the suite locally — exactly the people with a stale bin/wt or a Homebrew wt installed.

What changed

• findWtBinary → resolveWtBinary, returning (path, cleanup, error). Without -wt it builds into os.MkdirTemp and removes it afterwards; with -wt it uses the given path verbatim.
• The build runs with cmd.Dir set to the module root (go list -m -f {{.Dir}}), so invoking the runner from a subdirectory still builds the right package.
• main split into run() int so defer cleanup() runs on every exit path, including the failure exit.
• A -wt path that is missing or a directory is now a clear error up front, rather than an opaque exec failure repeated inside every scenario.
• The resolved binary is printed along with how it was obtained, so a surprising result is traceable to the artefact that produced it.

Test coverage

New e2e/run_test.go — the orchestrator had no unit tests:

• builds from the working tree, returns an absolute path, the result actually runs, cleanup removes it
• ignores a wt on PATH (decoy script; skipped on Windows)
• ignores a stale bin/wt — the other half of e2e: runner silently tests a stale or unrelated binary when --wt is not given #141
• rejects a missing -wt path rather than falling back
• rejects a directory as -wt
• uses an explicit -wt path verbatim

Red-green verified: restored the PATH fallback, confirmed TestResolveWtBinaryIgnoresWtOnPath failed; restored the ./bin/wt candidate scan, confirmed TestResolveWtBinaryIgnoresStaleBinArtifact failed with resolveWtBinary selected the stale bin/ artifact instead of building. Both restored.

Test plan

• go test ./... — all pass
• go vet ./... — clean
• gofmt -l . — clean
• go run e2e/run.go from a worktree with no bin/ — 251 passed, 0 failed, 13 skipped (previously 4 phantom failures in this exact setup)
• Codex review loop, 3 rounds → no actionable findings remaining

CI impact

None. .github/workflows/ci.yml already passes --wt=bin/wt / --wt=bin/wt.exe on every platform, so only the local default path changes behaviour.

[codecov]

Codecov Report

❌ Patch coverage is 52.94118% with 24 lines in your changes missing coverage. Please review.
✅ Project coverage is 43.14%. Comparing base (e241cf1) to head (8401242).

Files with missing lines	Patch %	Lines
e2e/run.go	52.94%	24 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #142      +/-   ##
==========================================
+ Coverage   42.48%   43.14%   +0.66%     
==========================================
  Files          34       34              
  Lines        3580     3597      +17     
==========================================
+ Hits         1521     1552      +31     
+ Misses       2059     2045      -14     

Files with missing lines	Coverage Δ
e2e/run.go	5.62% <52.94%> (+5.62%)	⬆️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.