Add intersect_vector16_inplace to enable vectorized inplace intersection between arrays.

[CharlesChen888]

I notice that in function array_container_intersection_inplace, there is no implementation of vectorized inplace intersection, and this slows down the computation.

We can simply use intersect_vector16 to do a non-inplace intersection, and replace the source array with the new one, but this requires frequent memory allocation and freeing. So I think a real inplace intersection is better.

Function intersect_vector16_inplace is modified from intersect_vector16, which uses __m128i tmp[2] for temporary storage of intersection result, to avoid overwritting original numbers in source array that haven't been used.

[lemire]

Running tests. Looks good at a glance!!!

[lemire]

GitHub complains about commented out code, make sure that you don't need to delete it.

[CharlesChen888]

I found this in fuzz/croaring_fuzzer.c. Looks like we need to comment out the first line.

int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size){
int LLVMFuzzerTestOneInput(const char *data, size_t size) {
    // We test that deserialization never fails.
    roaring_bitmap_t* bitmap = roaring_bitmap_portable_deserialize_safe(data, size);
    if(bitmap) {
        // The bitmap may not be usable if it does not follow the specification.
        roaring_bitmap_free(bitmap);
    }
    return 0;
}

[CharlesChen888]

@lemire I have rebased this branch to the latest commit, please rerun the tests.

[lemire]

I found this in fuzz/croaring_fuzzer.c. Looks like we need to comment out the first line.

This was a typo, I fixed it earlier.

[lemire]

Running tests.

[CharlesChen888]

@lemire Seems that fuzzers won't build. Have you got any idea why? Or does it matter?

[lemire]

This is a good PR. I am merging it. I will figure out the issue with the fuzzer later. It is unrelated to your code, I expect.