Fix the Qt4/Qt5 port that left the app unable to start, and modernise the template

[bgriffen]

Summary

The application did not start. CommonMPL.py — the module main.py imports — was still entirely Qt4, despite the earlier "updated to qt5 and traits6" commit, which only ported the standalone demo's private copy of the same editor. On the versions requirements.txt itself pins, importing it raises before anything renders:

mpl.use('Qt4Agg')                     -> ValueError, backend removed
matplotlib.backends.backend_qt4agg    -> removed in matplotlib 3.5
traitsui.qt4.editor                   -> removed in TraitsUI 8

This branch fixes that and the defects it was hiding, then adds the tests, CI and docs to keep it fixed.

What was broken

Area	Defect
CommonMPL.py	Qt4 imports throughout — app could not import, let alone run
main.py	All three marker handlers guarded on display_points, which nothing ever assigned, so they returned early forever. Had they not, they called wx.CallAfter — wx is not imported and not a dependency of a Qt app — and would have raised NameError
main.py	markercolor was set in __init__ but never declared as a trait, so it had no handler wiring and no UI control
main.py	_secondcalc_default returned an undefined SecondCalc
main.py	__init__ skipped HasTraits.__init__ and swallowed **kwargs
firstcalc.py	gradient/yintcept were Range(0.0, 5.0, 10.0) — default outside its own bounds. The editor clamps it to 5.0 while building the UI, firing _gradient_changed, whose figure.clear() destroyed the labelled axes before the window was shown
firstcalc.py	_rangeplotx_changed watched a non-existent trait and could not have run: a bare y statement plus undefined figure, xtmp, ytmp
firstcalc.py	plotbutton was labelled "Plot Me!" with no _plotbutton_fired handler
firstcalc.py	listoptions listed "Option 2" twice, so the menu only ever offered two choices
requirements.txt	Declared no Qt binding — neither TraitsUI nor matplotlib depends on one, so a clean install could not run the app
conda files	Two per-platform lock files pinning Python 3.5, Qt 4.8, traits 4.6 and matplotlib 1.5.1 to exact build hashes that no longer resolve, contradicting requirements.txt, and leaking absolute paths from the author's machine

What changed

Nine atomic commits:

1. chore — add .gitignore and MIT licence
2. fix — port the matplotlib editor from Qt4 to Qt5
3. fix — repair the dead marker controls in ApplicationMain
4. fix — correct invalid slider ranges and remove the broken handler in FirstCalc
5. refactor — reuse the shared editor in the standalone demo
6. build — modernise dependency declarations
7. test — add a headless regression suite
8. ci — run tests and lint on every push and pull request
9. docs — rewrite the README and refresh the screenshot

Two design points worth review:

• Import order in CommonMPL.py is load-bearing. pyface.qt is imported first so it resolves the Qt binding; matplotlib then reuses it rather than independently picking a different one and crashing on the first shared widget. The imports are fenced with # isort: off so a formatter cannot silently undo this.
• plot_xy() on ApplicationMain now owns the line. Tabs draw through it instead of rebuilding the figure by hand. That keeps the axis labels alive across replots and gives the marker controls a live handle on what is drawn. The duplicated, byte-identical _gradient_changed/_yintcept_changed bodies both collapse into it.

Verification

• 44 tests, each pinned to a specific defect above, passing on both a dev environment and a clean pip install -r requirements.txt.
• Confirmed the suite has teeth by reintroducing three original bugs; each fails the test written for it.
• ruff check . clean.
• main.py and standalonematplotlib.py both build their full widget trees offscreen.
• The README's "adding a tab" example was applied to a copy of the tree and run.

CI covers Python 3.9, 3.11 and 3.12 headlessly via Qt's offscreen plugin — no display or xvfb needed.

Please check before merging

• The MIT licence is my choice, not yours. The repo had none; swap or drop it if you want different terms.
• Deleted files: the two conda lock files, the empty root __init__.py, and both old screenshots (they predate the port and show the empty plot area the startup bug produced, replaced by one current capture).
• No packaging metadata was added deliberately — publishing top-level modules named main and CommonMPL to site-packages would collide with anything else doing the same. pyproject.toml carries pytest and ruff config only.