Turn the notebook into a walk-through of the method; render the math

[joeljose]

Closes #76

Notebook

It now mirrors docs/theory.md (20 cells, about 1,000 words of explanation, up from about 340), with runnable experiments on face.mp4:

1. Setup and input.
2. DTCWT sub-bands of a frame.
3. Phase against sub-pixel shift per level. It prints the slope, wavelength and half-turn shift: 4.1 / 6.0 / 11.2 / 19.6 / 43.3 px wavelengths.
4. One coefficient's phase over time, with its baseline and detail.
5. The width-mode gain G(f) against band mode. It prints the 0.20–8.59 Hz band and the 0.1 Hz resolution.
6. The full magnification, with switches for color_space and band_hz.
7. Before/after comparison.
8. The displacement limit on a synthetic ring: 2 px gives 0.95k with 1.7% ghost energy; 8 px gives 0.61k with 27.4%.

Executed top to bottom with jupyter nbconvert --execute in a fresh python:3.12-slim container with NumPy 2 preinstalled, as on Colab. There were no errors, all six figures rendered, and face_k3.avi was written.

Math rendering

• docs/theory.md and the notebook now use LaTeX ($...$, $$...$$) instead of plain-text formulas, so GitHub, Jupyter and Colab typeset them.
• Checked with GitHub's own Markdown API (POST /markdown): all 61 expressions in theory.md and all 26 in the README come back as math, unchanged. The 4 other $...$ matches in the README are shell code ($(id -u)) and correctly stay code.
• Two GitHub pitfalls found and avoided:
  • \, (thin space) loses its backslash, because Markdown treats it as an escape.
  • < gets double-escaped inside math and would show as <, so it's written as \lt.

README notebook description and CHANGELOG updated.