PERF: always slice when indexing on columns

[jbrockmendel]

• closes #xxxx
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

Touched on in #32779, this makes all* DataFrame indexing-on-columns do slicing instead of taking, so doesn't make copies.

* Actually there is a different path in _slice_take_blocks_ax0 that we go down if we self._is_single_block, can update that later if we decide this is something we want to do.

In [3]: dti = pd.date_range("2016-01-01", periods=10**5, freq="S")                                                                                                                                                  
In [4]: df = pd.DataFrame._from_arrays([dti]*10 + [dti - dti] * 10 + [dti.to_period("D")]*10, columns=range(30), index=range(len(dti)))                                                                             

In [8]: arr = np.arange(30)                                                                                                                                                                                         
In [9]: np.random.shuffle(arr)                                                                                                                                                                                      

In [10]: %timeit df[arr]
8.35 ms ± 64.1 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)    # <-- master                                                                                                                                                     
650 µs ± 52.5 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)   # <-- PR

The tradeoff is that we end up with less-consolidated results. I'm OK with that, but there may be downsides I'm not aware of, will wait for others to weigh in.

[jorisvandenbossche]

@jbrockmendel can you maybe open an issue for this? As your questions seems to be more broad than the actual PR, and to not mix that discussion with code review of the actual PR

[jbrockmendel]

i cant reproduce the test failures locally

[jbrockmendel]

Opened #33780, mothballing this.

[jreback]

this is quite old, happen to reopen if actively worked on.