[WIP]Perf[BMQ]: fast memcpy properties from continuous buffers

[678098]

We cannot not directly cast/assign a pointer even for simple types because alignment in a Blob is not guaranteed. Have to call memcpy with hopes that it will be optimized
TODO: check asm outputs

[678098]

int BlobUtil::readUpToNBytes(char*               buf,
                             const bdlbb::Blob&  blob,
                             const BlobPosition& start,
                             int                 length)
{
    if (BSLS_PERFORMANCEHINT_PREDICT_UNLIKELY(!isValidPos(blob, start))) {
        BSLS_PERFORMANCEHINT_UNLIKELY_HINT;
        return -1;  // RETURN
    }

    if (BSLS_PERFORMANCEHINT_PREDICT_UNLIKELY(length == 0)) {
        BSLS_PERFORMANCEHINT_UNLIKELY_HINT;
        return 0;  // RETURN
    }

    char*              outPosition = buf;
    const int          numBuffers  = blob.numDataBuffers();
    bmqu::BlobPosition cursor(start);

    while (length != 0 && (cursor.buffer() != numBuffers)) {
        const bdlbb::BlobBuffer& buffer   = blob.buffer(cursor.buffer());
        const int                buffSize = bufferSize(blob, cursor.buffer());

        const int toCopy = bsl::min(buffSize - cursor.byte(), length);
        bsl::memcpy(outPosition, buffer.data() + cursor.byte(), toCopy);

        outPosition += toCopy;
        length -= toCopy;

        cursor.setBuffer(cursor.buffer() + 1);
        cursor.setByte(0);
    }

    return static_cast<int>(outPosition - buf);
}

Short memcpy path should skip all these other operations