feat(python): extend existing fast range->Series init to lists of ranges in a Series

[alexander-beedie]

We currently have a range => Series fast-path; this extends the concept slightly to allow for easy creation of lists of ranges within a Series (previously would just dump the ranges in as unexpanded python objects - something for which there is no realistic use).

Example:

pl.Series( "a", [range(-2,1),range(3),range(2,8,2)] ) 

# shape: (3,)
# Series: 'a' [list]
# [
#     [-2, -1, 0]
#     [0, 1, 2]
#     [2, 4, 6]
# ]

pl.Series( "b", ((range(n),range(n+1),range(n+2)) for n in range(1,4)) ) 

# shape: (3,)
# Series: 'b' [list]
# [
#     [[0], [0, 1], [0, 1, 2]]
#     [[0, 1], [0, 1, 2], [0, 1, ... 3]]
#     [[0, 1, 2], [0, 1, ... 3], [0, 1, ... 4]]
# ]

Note:

The 1D case (range => Series) is fast; the 2D case (above) is currently "ok" as the number of distinct ranges gets large; each range is mediated by creation of an associated DataFrame inside pl.arange. Would be interesting if we could push some kind of sequence generation based on row number (like the above) into an expression...🤔

(FYI: I originally though pandas had some insanely fast method of doing this 100s of times faster than us, until I realised it was just loading the range objects "as-is" but its repr was unpacking a few rows of them when you eyeball the frame, hah! Almost spit out my tea when it first appeared to benchmark 586x faster - then reality reasserted itself :)

[ritchie46]

Anything over 100x difference is comparing apples to peaches is my rule of thumb. ;)