ISampler<T> performance tuning.

[colgreen]

Some ideas to consider:

Add an ISampler.Sample(out T value) method. This allows storing the sample result directly into an existing memory location, instead of returning a value that is subsequently moved to the mem location. This affects all PRNGs and 'numeric' samplers, i.e., e.g. ZigguratSampler.

In ZigguratSampler, we can avoid multiplying by the 'sign' double, with the following technique:

 ulong signBit =  (u & 0x400UL) << 53;
 SetSignBit(ref x, ref signBit);

 private static void SetSignBit(ref double x, ref ulong signBit)
 {
     Unsafe.As<double,ulong>(ref x) |= signBit;
 }

Initial performance tests showed a slowdown using .NET5, and a speedup using .NET6 (preview 3); but with .NET 6 being slower than .NET 5 to start with! Hence, I am leaving this open for now, and will revisit when a later .NET6 preview is out (or perhaps the final release).

[colgreen]

Done in commit 94c6b3b6cc8c927f92914dcfe0c3100e78c40cf4