[codex] Fix QE matdyn frequency token parsing

[singularitti]

Summary

Fixes #86 by parsing QE matdyn.x frequency rows as numeric tokens instead of splitting on whitespace and adjacent minus signs.

Root Cause

The previous parser used re.split(r"\s+|(?<!\s)(?=-)", line) to handle merged negative frequencies such as -2471.6424-2400.9590. That delimiter-based approach can split valid scientific notation at exponent signs, for example 1.0000E-03.

Changes

• Add a numeric token regex for frequency-line parsing.
• Replace delimiter splitting in FromQEOutput.read_frequency_file with number extraction.
• Add regression tests for merged negative frequencies and exponent signs.

Validation

• uv run python -m unittest tests.test_input_maker