ValidateError on grid length < 2 & variant cleanup

[kylecarow]

Summary

• A grid dimension with 0 or 1 points passed construction whenever Extrapolate::Enable
  wasn't selected, since the "at least 2 points" check only ran for that one setting.
  Every bundled strategy actually needs 2+ points to bracket a query regardless of
  extrapolation mode, so a 1-point dimension panicked on the first interpolate call
  instead (integer underflow in the lower-bracket search, or an out-of-bounds index
  right after it). Reproduced with Interp1D::new(array![5.0], array![10.0], strategy::Linear, Extrapolate::Error).unwrap().interpolate(&[5.0]).
• The check now runs unconditionally in InterpData::validate / InterpDataND::validate,
  turning that panic into a ValidateError::InsufficientGridPoints at construction time.
• Breaking: ValidateError variants renamed for consistency, no longer full
  sentences: ExtrapolateSelection -> InvalidExtrapolate, Monotonicity ->
  NonMonotonic. EmptyGrid is removed outright; InsufficientGridPoints now covers
  both the 0-point and 1-point cases with a single check.
• README.md's error list updated to match.

Test plan

• New regression test test_insufficient_grid_points covering the exact panic
  scenario above
• cargo build --all-features
• cargo test --all-features (69 unit tests, 15 doc tests, all passing)
• cargo clippy --all-features --all-targets (clean)
• cargo fmt --check