ENH: Generalize RangeIndex to support floats

[TomAugspurger]

Is your feature request related to a problem?

When working with geospatial data in xarray, it's common to have regularly-spaced, floating-point coordinates. Using a RangeIndex with floating-point start, stop, and step would cut down on the memory usage. https://github.com/gjoseph92/stackstac/blob/1c6553a5b6b2acbb89388bd8b1c2d280ea2f3bd5/stackstac/prepare.py#L412

Describe the solution you'd like

Let RangeIndex accept floating point start, stop, and step. Will need to change a handful of other methods to support the generalization (e.g. inferred_type, probably some of the hashtable stuff).

API breaking implications

None.

Describe alternatives you've considered

Reimplementing all of the index classes to support an "encoding" scheme, which would subsume range encoding, categorical encoding, ... But that's a ton of work :)

Additional context

>>> pd.RangeIndex(0.0, 1.0, 0.1)
RangeIndex(start=0.0, stop=1.0, step=0.1)  # same valuesas np.arange(0.0, 1.0, 0.1)

[marianna13]

Hi! In order to make this happen I think we should modify Python's built-in function range.

[jbrockmendel]

+1 on the idea.

FWIW A while back I tried to introduce a RangeEngine to allow for getting more code out of RangeIndex. Ran into a problem that cython doesn't recognize 'range' as a type, xref cython/cython#4040

[marianna13]

This modification of range function works with floats:

class float_range():
  def __init__(self, start=0, stop=None, step=1):
    self.start = start
    self.current = start - step
    self.stop = stop
    self.step = step
  
  def __iter__(self):
    return self

  def __next__(self):
    self.current += self.step
    if self.current < self.stop:
      return self.current
    raise StopIteration

Example:

rng = float_range(0, 2, 0.5)
for r in rng:
  print(r)

Output:

0.0
0.5
1.0
1.5

[kapiliyer]

take