Fixes numpy#29315 by making  return  consistently with
 instead of raising an IndexError. The fix:

1. Explicitly checks for empty arrays (size=0) in _quantile()
2. Returns NaN/NaT-filled array with correct shape and dtype
3. Maintains consistency with median behavior for empty inputs
4. Preserves all existing functionality for non-empty arrays

Handles all numeric, datetime and timedelta dtypes appropriately.

Fixes integer overflow in quantile calculation by converting integer arrays
to float64 before interpolation. This ensures:

1. Correct calculation for extreme values (e.g., [32767, -1] in int16)
2. Consistent results with median for integer inputs
3. No effect on floating-point or datetime types

The fix handles all integer, unsigned, and boolean types by safely casting
to float before interpolation operations while maintaining existing behavior
for other types. Matches median's behavior of returning float for integers.