PERF: Performance improvement value_counts for masked arrays

[phofl]

• closes #xxxx (Replace xxxx with the Github issue number)
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

ser = pd.Series([1, 2, pd.NA] + list(range(1_000_000)), dtype="Int64")
ser.value_counts(dropna=True/False)

# old dropna=False
# 99.6 ms ± 599 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

# new dropna=False
# 46.1 ms ± 349 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

# old dropna=True
# 138 ms ± 639 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

# new dropna=True
# 44.6 ms ± 439 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

data = np.array(list(range(1_000_000)))
Series(data, dtype="Int64")

# old Series(numpy_array, dtype="Int64")
# 45.8 ms ± 499 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

# new
# 63.8 µs ± 329 ns per loop (mean ± std. dev. of 7 runs, 10,000 loops each)

cc @jorisvandenbossche

[jorisvandenbossche]

Thanks, looks good!

[jorisvandenbossche]

I think we already have a benchmark for this in array.py (IntegerArray::time_from_integer_array), except that that one is with a tiny array and thus won't cover this aspect. But maybe add a version with a larger array in the existing benchmark? (or just make the array in that benchmark bigger)

[phofl]

Thx, forgot to check for series after I found nothing for value_counts. Increased the array size for the existing benchmark. 4 values is probably a bit small to see anything with overhead from other calls that don't depend on array size

[jorisvandenbossche]

Suggested change

	class ValueCountsEa:
	class ValueCountsEA:

? Small nitpick: not sure if we have some consistency around this, but if not, since we typically use EA as abbreviation, would keep it capitalized here

[phofl]

Done, will keep in mind for future prs