[RaphaelFakhri]

Fixes #145.

Problem

CompiledRunner transpiles a call to a user-supplied function as module.qualname, for example __main__.lower(...). Transpiler.evaluate() runs the code with the globals of celpy.evaluation, which only know the modules that celpy.evaluation imports. A function defined in any other module, such as __main__ or an application package, fails with:

CELEvalError: ('evaluation error', ..., ('undeclared reference to "name \'__main__\' is not defined" ...

The same expression works with InterpretedRunner.

Change

This revives the approach from #146 (closed because of failing checks), with these adjustments:

• Transpiler.__init__() collects the top-level package of each function's module from sys.modules. It maps _operator to operator the same way Phase1Transpiler.func_name() does.
• Transpiler.evaluate() executes the code with a copy of the celpy.evaluation globals plus those packages. Existing module globals take precedence, so a user package can't shadow a name that the transpiled code depends on.
• Type annotations satisfy mypy --strict.

Functions defined with lambda or nested inside another function still can't be referenced by name. That limitation is separate from this fix.

Testing

• New test test_transpile_function_from_other_module in tests/test_transpilation.py. It registers a function in a module that celpy.evaluation never imports and evaluates device.lower() == 'tablet' with CompiledRunner. The test fails without the change (NameError for the module name) and passes with it.
• pytest tests: 433 passed, 1 skipped, 100% coverage.
• pytest --doctest-modules src README.rst: 11 passed.
• behave --tags=@conformance --tags=~@wip, with both the interpreted and the compiled runner: 1127 scenarios passed, 0 failed.
• mypy --strict, ruff check src tools and ruff format --diff are clean.