Fix Random.NextSingle being able to return 1.0

[SimonCropp]

Bug

The Random.NextSingle polyfill (below net6) was (float) target.NextDouble(). A double just below 1.0 rounds up to exactly 1.0f when cast, which is outside the documented [0, 1) range. It is rare but real: new Random(0) returns 1.0f on call 5,800,140 on net48.

Fix

Sample again while the cast result is not below 1f. This is what the BCL does for its seeded implementation.

Test

RandomNextSingle_NeverReturnsOne takes 5,900,000 values from new Random(0) and asserts the maximum is below 1.0f and the minimum is not below 0.0f. It failed on net48 before the fix (maximum was 1), and passes on net462 and net48 (polyfill) and net10.0 (BCL).