It was libc's random, which is... Trash.
Now it use's xoshiro256 [a], which I hadn't heard of. (Not my area).  It seems somewhat similar to the MT, as it isn't cryptographically safe, but should give you a fast and fairly random set of integers back. It's also significantly faster than the MT.
[a] Followed by two asterisks. I give up trying to make them appear in HN's formatting.
Not my area either, but a little surprised why a PCG-family non-CSPRNG algorithm wasn’t used: https://www.pcg-random.org/posts/a-quick-look-at-xoshiro256....
PCG XSL RR 128/64 (MCG) could be used for 64-bit values (albeit with period 2^126 compared to xoroshiro256’s 2^256 − 1) with 128 bits. Half the state size than xoshiro256, and the performance might be comparable (would need benchmarking). LCG version if you want to trade PRNG quality for even more speed.