Add upsampling to the client

[TalweSingh]

This PR introduces upsampling to the client and adds test to check this functionality.

[shsms]

Didn't look into the resampling code.

[shsms]

Suggested change

	raise ValueError(
	f"Original period {original_period} must be positive non-zero"
	)
	raise ValueError(
	f"Original period {original_period} must be greater than zero"
	)

[shsms]

Suggested change

	# Validate periods are positive non-zero
	# Validate periods are greater than zero

[shsms]

Maybe like this?

Suggested change

	# Validate target period divides evenly into original period
	if original_period.total_seconds() % target_period.total_seconds() != 0:
	raise ValueError(
	f"Target period {target_period} must divide evenly into "
	f"original period {original_period}"
	)
	# Validate target period is an exact divisor of the original period
	if original_period.total_seconds() % target_period.total_seconds() != 0:
	raise ValueError(
	f"Target period {target_period} must be an exact divisor "
	f"of the original period {original_period}."
	)

[cwasicki]

Instead of the target period, you can pass in the requested timestamps. That simplifies the code below, allows arbitrarily sampled data and would align with what the user requests from the client. Namely they would request weather forecasts for specific datetime objects which are 15 min resampled. In this approach you first extract the target period (which could fail if it's not equidistant) and then re-construct the timestamps below for resampling.

[cwasicki]

Afterwards I think this whole function could be simplified roughly like the following:

First resampling

def resample(forecasts, validity_times, new_timestamps, axis=0):
    vts = np.array([t.timestamp() for t in validity_times])
    nts = np.array([t.timestamp() for t in new_timestamps])

    resampled = np.apply_along_axis(
        lambda arr: np.interp(nts, vts, arr),
        axis=axis, 
        arr=forecasts
    )

    return resampled

Then step shifting the solar feature indices (idxs):

def shift_idxs(arr, steps, idxs):
    # Modifies in place and bfills
    arr[steps:, :, idxs] = arr[:-steps, :, idxs]
    arr[:steps, :, idxs] = arr[steps:steps+1, :, idxs]
    return arr