Add error for overflowing nbits during PQ construction (#3833)

Summary:
Pull Request resolved: #3833

size_t expands to unsigned long int, so any left shift more than 31 means ksub overflows. So, we can add a check right before it is constructed in ProductQuantizer.

Ramil talked to Matthijs and the outcome is that 16 bits isn't really practical, but we can do the check for 24 to be safe.

Reviewed By: kuarora, mengdilin

Differential Revision: D62153881

fbshipit-source-id: 721db6bf6ad5dd5d336b4498f4750acc4059310c

faiss/impl/ProductQuantizer.cpp

@@ -61,6 +61,7 @@ void ProductQuantizer::set_derived_values() {
             "The dimension of the vector (d) should be a multiple of the number of subquantizers (M)");
     dsub = d / M;
     code_size = (nbits * M + 7) / 8;
+    FAISS_THROW_IF_MSG(nbits > 24, "nbits larger than 24 is not practical.");
     ksub = 1 << nbits;
     centroids.resize(d * ksub);
     verbose = false;