[DavisNT]

This is an improved version of #2478 - this preserves PRNG existing entropy when re-seeding.

This improves entropy of Dice and makes PRNG state recovery harder.
As a pleasant side effect this also makes the Dice app more like a physical dice - dependent on how it is shaken, but in a hard to predict manner.
Using PRNG output as one of the seeds preserves 32-bits of previous PRNG entropy when reseeding. For std::minstd_rand that means - its entire entropy.

This PR:

• Is almost free
  • Needs no additional RAM
  • Increases code size only by around 32 bytes
  • Only moves a few lines of code and adds only static_cast<uint32_t>(gen())
• Makes PRNG state recovery attacks incomparably harder. Recent demonstrations of LLM capabilities suggest such attacks might in a couple years be available to anyone using publicly available AI tools.
• Fixes the main drawback of Reseed Dice RNG on every Roll #2478 - discardation of the existing PRNG state/entropy during re-seeding. I.e. with this PR it should be much harder (especially if multiple independent people do turns with the Dice app) to fixate a certain output from PRNG by manipulating timing and position (or accelerometer saturation) of the watch.

TL;DR I think this could be a good addition to #2473 and a pretty cool thing on it's own as it makes the Dice app more like a real physical dice. 😎 🎲
This PR provides better security than #2478 at virtually no cost.

[github-actions]

Build size and comparison to main:

Section	Size	Difference
text	393264B	32B
data	936B	0B
bss	22920B	0B

Run in InfiniEmu