colblk: Bitmap, BitmapBuilder bug fixes

[jbowens]

Previously, Invert used slices.Grow to grow the b.words slice if necessary. If len(b.words) < cap(b.words), words in the unused capacity were not initialized to zero and were assumed to be zero. Most of the time this was okay, because Reset zeroed b.words up to len(b.words).

However if the BitmapBuilder was previously used to Finish fewer rows than were Set, it's possible Finish may have truncated b.words to exclude words with non-zero bits. Since Reset only reset words within len(b.words), the truncated words were left unzeroed.

This commit updates Invert to make no assumptions as to the value of words beyond len(b.words). It also removes the zeroing in Reset which obscured this bug in most cases, instead relying on Set, Finish and Invert to properly initialize memory as necessary as it's used.

Fix #4103.

[cockroach-teamcity]

This change is 

[RaduBerinde]

Nice catch! slices.Grow does append(s[:cap(s)], make([]E, n)...):len(s)] and from that I incorrectly assumed any new elements are zero - which is not true for elements between len(s) and cap(s). Plus the doc doesn't make any guarantees anyway.

Reviewed 2 of 2 files at r2, all commit messages.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @itsbilal and @jbowens)

---

sstable/colblk/bitmap_test.go line 160 at r1 (raw file):

		buildSize := size + rng.IntN(5)
		v := make([]bool, buildSize)
		for i := 0; i < len(v); i++ {

[nit] := range v

[jbowens]

Makes sense!

TFTR!

Reviewable status: 0 of 2 files reviewed, 1 unresolved discussion (waiting on @itsbilal and @RaduBerinde)