feat(python): support Series init from generator

[alexander-beedie]

Closes #5395.

---

Currently Series init requires full materialisation of generators/iterables, up-front. As we don't have the same costs relating to chunked-append as pandas, we can take the more memory-friendly approach of incremental materialisation followed by a final call to rechunk.

Before: (don't actually do this - much more efficient to init 'range' directly)

import polars as pl

def gen(n):
    yield from range(n)

pl.Series("s", gen(5), dtype=pl.Int8)

# ValueError: Series constructor not called properly. 
#  Got <generator object gen at 0x17d2fc200>.

After:

pl.Series("s", gen(5), dtype=pl.Int8)

# shape: (5,)
# Series: 's' [i8]
# [
#     0
#     1
#     2
#     3
#     4
# ]

Also:

Ensure Series generated from python range object still respect the given integer dtype, if provided (was being ignored; always i64).

Follow-up:

Offer generator support for DataFrame init; will take care of that later, in a separate PR.