Stop reseeding RNG on every solve call (fixes #464)

[ChrisRackauckas-Claude]

Summary

This PR implements the change requested in #464 - stopping the automatic reseeding of the RNG on every call to solve.

Problem

Previously, JumpProcesses would reseed the RNG with rand(UInt64) on every call to solve when:

• seed === nothing (no explicit seed provided)
• alias_jump == false (i.e., when not on the main thread, or when deepcopying the problem)

This led to non-reproducible results because it didn't respect the user's RNG choice.

Solution

• Modified resetted_jump_problem in src/solve.jl to only reseed when an explicit seed is provided
• Modified DiffEqBase.__init in src/SSA_stepper.jl similarly
• Updated test/ensemble_uniqueness.jl to use prob_func with explicit seeds for each trajectory

Breaking Change

This is a breaking change as noted in #464. For EnsembleProblems that require different trajectories, users should now use prob_func to set unique seeds for each trajectory. This gives users full control over reproducibility.

Example of the new pattern for EnsembleProblems:

using JumpProcesses, StableRNGs

function make_seeded_prob_func(dprob, aggregator, jumps, base_rng)
    return function prob_func(prob, i, repeat)
        seed = rand(base_rng, UInt64)
        JumpProblem(dprob, aggregator, jumps...; rng = StableRNG(seed))
    end
end

# Create JumpProblem
jump_prob = JumpProblem(dprob, Direct(), j1, j2)

# Create EnsembleProblem with unique RNG per trajectory
ensemble_rng = StableRNG(99999)  # seeded for reproducibility
ensemble_prob = EnsembleProblem(jump_prob;
    prob_func = make_seeded_prob_func(dprob, Direct(), (j1, j2), ensemble_rng))
sol = solve(ensemble_prob, SSAStepper(), trajectories = 10)

Test plan

• All existing tests pass (verified locally)
• Updated test/ensemble_uniqueness.jl to demonstrate the new pattern
• CI should pass

Fixes #464

cc @ChrisRackauckas

🤖 Generated with Claude Code