[ENH] Parallel PAA & SAX transform

[baraline]

Describe the feature or idea you want to propose

Sax transform currently have n_jobs but use it only for inverse transform. We should also use it for the transform step.

Describe your proposed solution

We should introduce numba parallel replacements for _get_sax_symbols(self, X_paa) and also introduce n_jobs to PAA.

These functions are fairly simple to parallelize. For exemple the sax symbols (replace uint8 with appropriate dtype):

@njit(fastmath=True, cache=True, parallel=True)
def _get_sax_symbols(X, breakpoints):
    n_cases, n_channels, n_timepoints = X.shape
    X_new = np.zeros((n_cases, n_channels, n_timepoints), dtype=np.uint8)
    n_break = breakpoints.shape[0] - 1
    for i_x in prange(n_cases):
        for i_c in prange(n_channels):
            for i_b in prange(n_break):
                mask = np.where(
                    (X[i_x,i_c] >= breakpoints[i_b]) & (X[i_x,i_c] < breakpoints[i_b + 1])
                )[0]
                X_new[i_x, i_c, mask] += np.uint8(i_b)

    return X_new

Describe alternatives you've considered, if relevant

No response

Additional context

No response