[Go] Add test coverage for the rand package

[tauptlab]

Closes the TODO at the top of rand.go asking for test coverage of the exported noise-generating functions. The package only had a test for Boolean's bit-shifting before this.

The tests are in three layers:

• Byte-stream tests swap randBuf for a bytes.NewReader and check exact outputs. The targeted edge cases live here: I63n's rejection loop at (MaxInt64/n)*n, the sign-bit mask, Geometric's leading-zero counting, and Uniform's r==0 fallback. The fallback can't be reached from crypto/rand at any realistic call count, but it is reachable by feeding 128 zero bytes (forces Geometric() >= 1024 so math.Pow(2,k) overflows to +Inf).
• Range tests verify that outputs sit in the documented range, 10k samples each.
• Moment tests sample 1M times and compare mean and variance against population values, with 5-sigma tolerances derived from each distribution's variance and 4th central moment. The helper skips the variance check on Bernoulli-shaped distributions where mu_4 == sigma^4.

TestNormal_FourthMoment is a separate explicit check because Normal and Laplace at the same variance both pass the regular moment test - they only differ at mu_4 (3 vs 6). The library uses both for the Gaussian and Laplace mechanisms.

TestConcurrent stresses the package locks with 8 goroutines, plus a Boolean mean check so it has signal even without -race. testing.Short() skips the moment tests.

rand.go

• Dropped the test-coverage TODO.
• Added a one-line comment above randBuf saying it must wrap crypto/rand.

Two pre-existing edge cases noticed while writing the tests:

1. I63n doesn't validate n. n=0 panics with divide-by-zero, n<0 returns values in [0, |n|) instead of failing. Added a TODO; I'll send a follow-up PR.
2. randSource.Int63 returns math.MinInt64 once per 2^64 calls on the byte stream 0x8000000000000000 (since -math.MinInt64 == math.MinInt64 in two's complement). Added a Note in code rather than fixing - any fix would change the output mapping for ~half of all byte sequences, which seems disproportionate for a 2^-64 contract violation that's never been observed.

Testing

• go test ./go/... clean.
• -count=10 clean.
• -short correctly skips the moment tests.
• Couldn't run -race locally - no gcc on this Windows machine.