Fix dielectric attenuation, input validation, and header rebuilds

[osaether]

The calculator could report excessive dielectric attenuation and successfully calculate a different conductor configuration after rejecting an invalid conductor. Malformed numeric inputs could also produce NaN results or silently change frequency (30MHz became 30 Hz).

This change corrects the dielectric filling factor and attenuation coefficient, aborts on conductor validation errors, and requires finite numeric values with no trailing text. Mesh counts must be integers within their supported ranges. It also adds generated header dependencies to the Makefile and corrects the dB/m-to-Np/m explanation for gamma.

Regression tests build and run the C executable in temporary directories, cover all five findings, and check valid-input and lossless-substrate behavior. CI now runs the full test suite, and the README documents the corrected formula and input requirements.

Validation:

• The initial regression suite reproduced all five findings with 11 failing cases before fixes.
• python3 -m pytest -q: 56 passed after fixes.
• make -B: full rebuild succeeded.
• make check-z0: agreement within 0.03% of the Hammerstad-Jensen reference.
• git diff --check: passed.