fix(craft): improve multiplet quantitation accuracy

[nmrtist]

Summary

Improve CRAFT multiplet quantitation accuracy for Bruker FIDs with digital-filter group delay and non-Lorentzian signal shapes.

Problem

CRAFT treated retained Bruker samples as if their time coordinates still started at the original raw point index. This reintroduced the removed digital-filter delay, causing amplitude extrapolation error and a frequency‑dependent first‑order phase ramp.

The default limit of seven model components per fit window was also insufficient for broader, non‑Lorentzian multiplets. On the ethylbenzene ¹H reference data, the methyl‑to‑methylene ratio was 1.226889 instead of the expected 1.5.

Changes

• Use the physical FID time origin after subtracting the acquisition group delay during:
  • band mixing
  • time‑domain fitting
  • residual reconstruction
• Increase the default model components limit per fit window from 7 to 15.
• Add regression coverage for:
  • fractional group‑delay amplitude and phase recovery
  • multiplet quantitation requiring more than seven model terms
• Update the English and Simplified Chinese CRAFT documentation.

Results

For the ethylbenzene ¹H reference regions:

• Before: 1.226889
• After: 1.511513
• Expected: 1.5
• Relative error: reduced from 18.21% to 0.77%

CRAFT model components remain mathematical damped‑exponential terms and do not necessarily correspond one‑to‑one with visible multiplet lines.

Validation

• npm run build from docs/
• cargo pr-check

All checks passed.