fix(cli): wheels new emits app-relative paths in 'create' log

[bpamiri]

Summary

`wheels new ` printed misleading filesystem paths in its per-file `create` log lines:

```
$ wheels new blog
create blog/.gitignore
create blog/app/
create blog/snippets/
create blog/BoxJSON.txt ← actual: blog/app/snippets/BoxJSON.txt
create blog/ControllerContent.txt ← actual: blog/app/snippets/ControllerContent.txt
...
create blog/migrator/ ← actual: blog/app/migrator/
create blog/Model.cfc ← actual: blog/app/models/Model.cfc
create blog/README.md ← actually multiple READMEs in subdirs, all flattened to root
create blog/README.md ← (printed 3+ times)
```

Files landed correctly under `blog/app/models/`, `blog/app/snippets/`, etc. on disk — only the log was wrong. But it was misleading enough to make the harness's Phase 2 grep miss the path layout entirely (not just because of ANSI escapes — see below).

Root cause

`copyTemplateDir()` computed:

```cfm
var relativePath = arguments.appName & replace(targetPath, arguments.targetDir, "");
```

— but `targetDir` is the CURRENT recursion's target, not the project root. Each recursion level stripped more of the path. A file at `/app/models/Model.cfc` had `/app/models` stripped (the deepest recursion's targetDir), leaving the log line `/Model.cfc`.

Multiple READMEs (one per template subdir) collided into the same logged path for the same reason.

Fix

Added `rootTargetDir` parameter to `copyTemplateDir`, defaulting to `targetDir` on the initial call. Passed unchanged through recursion. Compute `relativePath` relative to root, not the current level. The initial caller in `scaffoldNewApp` passes `targetDir` as both arguments to anchor the recursion.

Also removed a duplicate `create /db/` line from `configureSQLiteDatabase` — the template copy already emits it when the template ships an empty `db/` dir.

Harness Phase 2 was silently false-PASSing

Two separate bugs in the harness's Phase 2 hid this issue for prior runs:

1. ANSI color escapes broke the grep. Both the F3 dup-line check and the misplaced-path check ran greps with `^[[:space:]]*create` against the raw log. But `wheels new` emits ANSI color escapes (`\x1b[...m`) at the start AND end of every line, so the regex never matched and the greps returned zero hits regardless of bug state. Strip ANSI to a plain log file first.

2. Multi-line `DUP_COUNT`. The dup-count computation used `grep -cv '^$' ... || echo 0`. On zero matches, `grep -c` emits "0" to stdout AND exits 1, so the `|| echo 0` fallback ALSO ran, producing "0\n0" — which broke `[ "$DUP_COUNT" -eq 0 ]` with "integer expression expected." Replaced with the `printf | grep | wc -l | tr -d ' '` pattern already used elsewhere in this script.

Result

Harness now reports: 43 pass / 0 fail / 2 skip / 45 total. Phase 2 reports both:

```
✓ F3: no duplicate 'create' lines in wheels new output
✓ wheels new prints paths under proper app/ public/ subdirectories
```

Only remaining tracked SKIP is #2332 (browser install) which needs real Playwright integration work, not a small fix.

Test plan

• `bash tools/test-cli-local.sh` — 448 pass / 3 fail / 0 error. Three failures are pre-existing Doctor Service issues (packages: doctor mixin-collision scan is regex-based and may miss metadata-driven mixins #2260) unrelated.
• `bash tools/test-onboarding.sh` Phase 2 fully PASSes.
• Manual probe confirms `create /app/models/Model.cfc` instead of `/Model.cfc`, and no duplicate README lines.

Closes #2328