Update demo video with bayonet selections and model picks

[altendky]

The README links to the older demonstration. Update it to the new 3:48 video, with narration focused on modeling the nozzle's bayonet mount and using relationships to avoid offset faces, off-center bores, and end faces that are not square to their cylinder axes.

The capture now starts from the bare nozzle scan, paints representative wall, rim, and slope selections, and creates fits and a parallel-plane relationship through actual Model picks. Prepared remaining features retain the exact filmed selections and fits. Shorten setup and evaluation waits, preserve completed interactions, and record playback rates, edits, and checkpoints in the manifest. Rendered media stays outside the repository.

Increase the default least-squares iteration budget from 100 to 300 so the nozzle recipe's offset cylinder converges, with a regression covering its convergence and the completed recipe.

Validation:

• Python feature-graph and solver checks, including the nozzle convergence regression; broader Python checks completed with the CLI workload tests passing in isolation.
• Browser syntax/unit checks: 274 passed; affected Chrome integration tests: 33 passed.
• Pre-commit checks and JavaScript syntax checks passed.
• Full video/audio decode, output/source hashes, captions, and preservation of filmed inputs verified.

Known limitation: the residual scene uses the existing all-fit scale and individual fit results. It does not show full-scan or legacy joint-fit residuals, and the shared scale can make the colors look pale.