Test tools force multimer's source with MULTIMER_SOURCE

[bdbarnett]

Fixes #150.

pydevices 0.6 reads MULTIMER_SOURCE and ignores MULTIMER_BACKEND, and multimer.auto no longer exists, so a forced-backend run either skipped as "unavailable" or tested the default source.

Now the preload (renamed tools/multimer_source_preload.py), the wrapper's --multimer-source and lv_timer_test_kit.py --source set MULTIMER_SOURCE, make multimer pick its source before the script runs, and check that the source in use is the one asked for. That check matters because a forced source that can't start falls back to none without raising. example_test_kit.py forwards MULTIMER_SOURCE and refuses to run when MULTIMER_BACKEND is set. The wrapper's no-thread quit schedule also imported multimer.auto; it uses multimer now. tools/timing_bench keeps MULTIMER_BACKEND on purpose, as the issue says.

What I ran, from a checkout at sibling depth so --source-workspace finds pydevices and lvgl-bindings:

• tests/test_multimer_source.py: 5 tests OK. The preload tests run it under CPython for pending, none and wasm (unavailable, exit 3). On main they fail because the preload that forces a source doesn't exist.
• lv_timer_test_kit.py --source signal: micropython signal/bytecode, ok, cpython-venv signal/bytecode, ok. --source pending: cpython-venv pending/bytecode, ok, micropython unavailable. --source none: micropython none/idle, ok.
• MULTIMER_SOURCE=pending example_test_kit.py --only-example lv_test_timer --only-interpreter cpython-venv: SDLDisplay, ok. With MULTIMER_BACKEND=sdl2 it stops with "MULTIMER_BACKEND is retired; set MULTIMER_SOURCE ...".
• ruff check lib tests tools scripts passes.

Not run: the Windows .exe interpreters and CircuitPython.