[lemire]

Fixes #592.

bitmapContainer.NextUnsetBit shifted the word right before inverting it, so the zeros shifted in at the top turned into ones and were read as unset bits. Whenever the rest of the word was all ones, it returned the first bit of the next word, even if that bit was set. The bitmap container's unset iterator (Unset, UnsetIterator, including AdvanceIfNeeded) uses this function, so it yielded values that are in the bitmap:

d := roaring.NewBitmap()
d.AddMany([]uint32{0})
d.AddRange(2, 4098)
for v := range roaring.Unset(d, 2, 4099) { fmt.Println(v) } // 64 128 192 ... instead of 4098 4099

The fix inverts first and then shifts, as nextAbsentValue already does (79ebb60).

The NextAbsentValue/PreviousAbsentValue part of #592 was already fixed in v2.30.0. The report's output matches v2.29.0.

Tests:

• TestBitmapContainerNextUnsetBitAcrossWords: NextUnsetBit on runs of ones that cross word boundaries, and on a full container.
• TestUnsetBitmapContainer: the reproductions from NextAbsentValue and PreviousAbsentValue answer with a value that is present in a bitmap container #592, checked against a brute-force reference over several ranges, plus AdvanceIfNeeded.

Both tests fail without the fix. go test ./... passes.