[MNT] Fix RDST CI failure and slow down

[baraline]

Reference Issues/PRs

Part1 to Fixes #2071

What does this implement/fix? Explain your changes.

Remove distance argument from RDST, uniformize dtypes, use of strides (view of the original array), add new random generator API instead of old one.

Current results on this branch (original train test split) :

Name	Acc	time
ACSF1	0.92	8.253568
Adiac	0.7161125319693095	4.220171
AllGestureWiimoteX	0.7242857142857143	11.283452
AllGestureWiimoteY	0.7157142857142857	15.981344
AllGestureWiimoteZ	0.6757142857142857	13.106202
ArrowHead	0.8571428571428571	1.481234
ArticularyWordRecognition	0.9966666666666667	16.601666
AtrialFibrillation	0.2	1.490244
BME	0.9933333333333333	0.842719
BasicMotions	1.0	1.438721
Beef	0.8333333333333334	1.08501
BeetleFly	0.95	0.96993
BirdChicken	0.9	1.009865

Current results on main (original train test split) :

Name	Acc	time
ACSF1	0.93	33.765473
Adiac	0.7340153452685422	12.60369
AllGestureWiimoteX	0.7128571428571429	38.506022
AllGestureWiimoteY	0.71	38.440987
AllGestureWiimoteZ	0.69	38.416986
ArrowHead	0.8628571428571429	4.390125
ArticularyWordRecognition	0.9933333333333333	33.320754
AtrialFibrillation	0.2	3.248225
BME	0.9933333333333333	2.330296
BasicMotions	1.0	2.900055
Beef	0.8666666666666667	2.638048
BeetleFly	0.95	2.430503
BirdChicken	0.9	2.444238