Fixed coverity issue in Philox RNG engine

[ElenaTyuleneva]

What introduces this PR:

1. Fix for the incorrect counter incrementing in case w is not the recommended value (w!=32 and w!=64)
2. discard_overflow_test was extended: previously only the first chunk of the counter was overflowing, now introduced the high-level loop, which iterates through the other positions of the counter(2nd and 3rd) and also checks for the overflow.
3. Introduced a new unit test - counter_management_test.
   • The testing is based on comparing the work of two different methods of the engine - simple set_counter and more complicated increase_counter_internal(unsigned long long __z) which is called by discard
   • The test discards the engine by the increment which takes at least 2 chunks of the counter type(2*w) - this way increase_counter_internal will take __z that should be placed properly in the 1st chunk and propagated to the 2nd one. The right calculations are checked by referenceEngine which is based on simpler set_counter function.

[aelizaro]

Suggested change

	unsigned long long counter_increment = increment / 4;
	unsigned long long counter_increment = increment / Engine::word_count;

[ElenaTyuleneva]

Thanks for catching that! Fixed + all unit tests are also extended for Philox 2x*(except for set_counter_conformance_test as it expects 1955073260 and 3409172418970261260 numbers to be generated as 10000th)

[aelizaro]

Looks good to me, thank you!