[tashakaggarwal]

Description of proposed changes

Several places validate numeric input with isinstance(x, int) / isinstance(x, (int, float)).
NumPy integers (np.int64, np.int32) and np.float32 are not instances of int/float, so
they are handled inconsistently compared with the equivalent Python numbers:

• Figure.text(angle=np.int64(30)) silently ignores the angle (no error, no +a30 in -F).
  This is the worst case because there is no feedback.
• Pattern(np.int64(15)) raises GMTValueError: Invalid pattern number.
• binstats(statistic="quantile", quantile_value=np.int64(50)) (or np.float32) raises
  GMTTypeError.

This PR checks against numbers.Real / numbers.Integral instead, which NumPy scalars register
with. Booleans are explicitly rejected for Pattern.pattern and binstats(quantile_value), since
True is an int in Python and previously produced Pattern(True) -> 'p' (a pattern with no
number).

Tests added for all three.

[seisman]

Thanks for the PR. I agree that it's more flexible to accept numpy scalars, but I feel we should have a more general way to handle it.