fitcheb ergonomics / seam cleanup (follow-ups from #2)

[ibell]

Tracking follow-ups observed during the work on #2. Each is a small, focused item — splitting so the fixes stay reviewable.

1. Relative default paths are working-directory-dependent foot-guns

fastcheb.cpp:31-33 defines:

```cpp
extern const std::filesystem::path teqp_datapath{ "../externals/CoolProp" };
extern const std::filesystem::path output_prefix{"../output/"};
extern const std::filesystem::path check_destination{"../outputcheck/"};
```

These only resolve correctly when fitcheb is run from build/. The README has already documented this as a quirk ("The code expects that the working directory when running is a subfolder of this code"), but with -d/-o/-c now fully wired, the defaults should be anchored to argv[0]'s directory (or the repo root) instead of CWD. One-line fix with std::filesystem::canonical at startup.

2. Delete inject_superancillary.py in CoolProp

The script at externals/CoolProp/dev/scripts/inject_superancillary.py is obsoleted by fitcheb inject (#2). It also has a hardcoded wrong path (~/Documents/Code/fastchebpure/output) and no hash check. Needs a separate PR against the CoolProp repo, not this one.

3. No CLI / subcommand tests

The -d-ignored-by-fit/check bug fixed in #2 would have been caught by a minimal smoke test that runs fitcheb fit -f Argon -d /tmp/isolated-tree and asserts it actually reads from /tmp/isolated-tree. Worth adding a small Catch2 or shell-level test harness for the three subcommands — at least covering inject's hash-guard, idempotency, and atomicity paths, which are the places bugs would silently corrupt fluid JSONs.

4. --skip help string renders oddly

fitcheb --help prints:

```
-s,--skip TEXT [[Air,SES36]] ...
```

Cosmetic — CLI11's capture_default_str() formatting on a std::vector<std::string> default wraps the list in extra brackets. Either switch to a single comma-joined string or override the default-str formatter.

5. README quirk note can go away once #1 lands

The "The code expects that the working directory when running is a subfolder of this code" note in README.md exists only because of #1 above. Delete the note at the same time.

---

Context: see #2 for the inject subcommand work these follow-ups fall out of.