Speedup approx entropy

[Scott-Simmons]

np.max(np.abs(x_re[:, np.newaxis] - x_re[np.newaxis, :]), axis=2) <= r was identified as a hotspot on the scalene profiler.*

This PR speeds this up.

Here is a snippet of before optimisations made:

Here is after changes made:

Does not change much... so this may be a red herring hotspot. Profiling the individual line change showed a massive speedup, but overall it harms performance... leaving in draft for now

Reproduce profiling results with:

from tsfresh.examples import robot_execution_failures
from tsfresh import extract_features
from tsfresh.utilities.dataframe_functions import impute
import pandas as pd
robot_execution_failures.download_robot_execution_failures()
df, y = robot_execution_failures.load_robot_execution_failures(
)
df_large = pd.concat([df]*100, ignore_index=True)
X = extract_features(
    df_large,
    column_id='id',
    column_sort='time',
    impute_function=impute,
    disable_progressbar=True,
)

scalene --cpu --profile --multiprocessing profile_tsfresh.py

Scalene profiler appears to be more informative than other python profiling tooIs (that I know of)