Fix thread issue with UUID generation

[svwilliams]

The original Boost random UUID generator is thread-safe, but it was very slow. Some time ago, I replaced the generator code with equivalent logic using the STL random library to improve performance. But had the downside of not being thread-safe. This manifests itself occasionally with different constraints having the same UUID.

This PR includes:

• A mutex lock on the random number generator to handle calls from multiple threads correctly.
• A slight refactor of how the random bits are copied into the UUID bytes in an attempt to minimize the lock time.

[svwilliams]

Please pay extra attention to this refactor.

My interpretation of the original code is that a random number of N bytes is generated. Inside the loop, each random byte is copied into the next open slot in the UUID. When you run out of random bytes, generate N more bytes.

The refactor relies on the uint8_t type in the UUID and the uint64_t type generated by the random number generator being the expected sizes.

That seems like a strong guarantee in C++:

unsigned integer type with width of exactly 8, 16, 32 and 64 bits respectively
(provided only if the implementation directly supports the type)

https://en.cppreference.com/w/cpp/types/integer

[needphpsql]

looks good to me.