Raise test coverage to 99% and fix a constant-value regression

[saudzahirr]

Coverage had fallen to 76% because most of the library was never exercised: umath was at 54% with nearly every trigonometric, hyperbolic and special function untested, and the operator overloads, the distribution constructors and the printed reports were largely untouched. The new tests take the package to 99%.

• test_umath_functions.py parametrizes all 39 unary functions plus the stepwise, power and binary ones, checking each twice: once with a plain float and once with an uncertain input. Reference values come from the standard library, so a wrong derivative shows up as a mismatch instead of agreeing with itself. umath goes from 54% to 100%.
• test_uncertain_function.py covers every operator, the derivative accessors, and all four combinations of error_components.
• test_method_of_moments.py covers the standardization helpers, the argument checks, the printed report and raw2central.
• test_distributions.py covers all eleven constructors against their closed-form mean and variance, plus every parameter-validation path.
• test_uncertain_variable.py gains the setters and the plotting paths.

Writing them surfaced a real regression from the FORTRAN port: a constant UncertainFunction has no input variables, so the kernel was handed zero-length arrays and f2py raised

ValueError: unexpected array size: new_size=1, got array with arr_size=0

The Python loops it replaced simply iterated zero times, so str(UncertainFunction(4.0)) used to work and had stopped. rawmoment, variance_components and soerp_numeric now short-circuit when there are no variables, which is also what to_uncertain_func needs for a plain number.

Coverage is now reported against src/soerp rather than site-packages. The package carries a compiled extension so it cannot be imported from the source tree, and without [tool.coverage.paths] the results do not line up with the repository. matplotlib joins the test group so the optional plotting path is exercised rather than skipped.

Two small things left uncovered deliberately: the PackageNotFoundError fallback in init, which needs the package reloaded to reach, and the defensive else branches in the error_components report, which cannot be reached because the dictionaries are built from the same variable list they are then looked up in.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!