Update usage of random number generators

[jhump]

Go 1.22 included a new "math/rand/v2" package (see the blog post for more details). The most exciting thing about this is a new global RNG that uses per-thread seeds and is lock-free. And it turns out that this approach is indeed much better than using a mutex with a *rand.Rand.

But for random numbers that don't need to be thread-safe, the Go 1 generator in "math/rand" is still the fastest.

So this PR removes internal.NewLockedRand() and updates all callers to instead use the global RNG in "math/rand/v2".

---

You can see the code I used to benchmark the RNGs in 43b8cd0. We only use a small number of operations of the RNG interface, so I wrote a benchmark for each:

• For non-thread-safe RNGs, we only use Uint64 and Shuffle methods.
• For thread-safe RNGs, we only use Intn (called IntN in the new "math/rand/v2" package) and Float64.
  For the thread-safe benchmarks, we actually spin up multiple goroutines (based on GOMAXPROCS), so that we can also see the costs of lock contention.

For all of them, I tested the existing implementation which uses the Go 1 "math/rand" generator. And I also tested using all three of the RNGs in "math/rand/v2": the PCG and ChaCha8 algorithms as well as its global RNG (which uses the runtime's internal, lock-free RNG).

In these benchmarks, the Go 1 generator in "math/rand" maintains a reasonable lead for the non-thread-safe operations. But for thread-safety, the global RNG in "math/rand/v2" wins handily.

goos: darwin
goarch: arm64
pkg: github.com/bufbuild/httplb/internal
cpu: Apple M1 Max
BenchmarkRand_Uint64_rand
BenchmarkRand_Uint64_rand-10                            406894315            2.849 ns/op
BenchmarkRand_Uint64_randv2pcg
BenchmarkRand_Uint64_randv2pcg-10                       292239162            4.061 ns/op
BenchmarkRand_Uint64_randv2chacha8
BenchmarkRand_Uint64_randv2chacha8-10                   257501427            4.613 ns/op
BenchmarkRand_Uint64_randv2global
BenchmarkRand_Uint64_randv2global-10                    226841599            5.294 ns/op
BenchmarkRand_Shuffle_rand
BenchmarkRand_Shuffle_rand-10                             4805521            248.8 ns/op
BenchmarkRand_Shuffle_randv2pcg
BenchmarkRand_Shuffle_randv2pcg-10                        4451755            267.9 ns/op
BenchmarkRand_Shuffle_randv2chacha8
BenchmarkRand_Shuffle_randv2chacha8-10                    3403008            349.9 ns/op
BenchmarkRand_Shuffle_randv2global
BenchmarkRand_Shuffle_randv2global-10                     3106603            382.8 ns/op
BenchmarkRand_ConcurrentIntN_rand
BenchmarkRand_ConcurrentIntN_rand-10                      1214431            969.0 ns/op
BenchmarkRand_ConcurrentIntN_randv2pcg
BenchmarkRand_ConcurrentIntN_randv2pcg-10                 1230388            951.7 ns/op
BenchmarkRand_ConcurrentIntN_randv2chacha8
BenchmarkRand_ConcurrentIntN_randv2chacha8-10             1000000             1027 ns/op
BenchmarkRand_ConcurrentIntN_randv2global
BenchmarkRand_ConcurrentIntN_randv2global-10             39565887            88.69 ns/op
BenchmarkRand_ConcurrentFloat64_rand
BenchmarkRand_ConcurrentFloat64_rand-10                   1270753            947.7 ns/op
BenchmarkRand_ConcurrentFloat64_randv2pcg
BenchmarkRand_ConcurrentFloat64_randv2pcg-10              1279284            928.6 ns/op
BenchmarkRand_ConcurrentFloat64_randv2chacha8
BenchmarkRand_ConcurrentFloat64_randv2chacha8-10          1272981            948.7 ns/op
BenchmarkRand_ConcurrentFloat64_randv2global
BenchmarkRand_ConcurrentFloat64_randv2global-10         148186993            7.944 ns/op

[ghost]

Nice.