Add LogExpFunctions 1.0 to compat

[ChrisRackauckas-Claude]

Please ignore until reviewed by @ChrisRackauckas.

Supersedes #75 (dependabot). This is the same compat bump (LogExpFunctions = "0.3, 1.0") plus a verification record.

Summary

LogExpFunctions v1.0.0's only listed breaking change is a docstring relaxation for log2mexp (JuliaStats/LogExpFunctions.jl#103). PoissonRandom does not use log2mexp.

The only LogExpFunctions API consumed here is log1pmx at src/PoissonRandom.jl:114. The v1.0 release added a Float32 implementation and improved accuracy for log1pmx, but the signature and behavior on existing inputs are unchanged. No code changes are required — only the compat bound.

Verification

Ran the full test suite (Aqua, JET static analysis, ExplicitImports, count_rand/ad_rand/pois_rand statistical samplers, BigFloat support, PassthroughRNG dispatch, AllocCheck zero-allocation tests) against LogExpFunctions 1.0.0 on:

• Julia 1.10.11 — all tests pass
• Julia 1.11.9 — all tests pass
• Julia 1.12.6 — all tests pass

Also spot-checked log1pmx directly with Float64, Float32, and BigFloat inputs against LogExpFunctions 1.0.0 — values match expectations.

Test plan

• CI passes on Julia 1.10, 1.11, 1.12 with LogExpFunctions 1.0.0 resolved
• CI also passes with LogExpFunctions 0.3.x (lower bound) still resolvable

🤖 Generated with Claude Code

[ChrisRackauckas-Claude]

CI status on this PR:

• Julia 1 (release, 1.12) and lts (1.10) — pass on ubuntu / macos / windows
• Julia pre (1.13.0-rc1) — fails, but this is a pre-existing failure also present on master (JET has no Julia 1.13 release yet)

Note: the matching dependabot PR (#75) shows all test jobs failing, which is not caused by the LogExpFunctions bump. It's the force_latest_compatible_version flag in julia-actions/julia-runtest — that flag is auto-enabled for dependabot PRs, and forces JET to its latest tagged version (0.11). JET 0.11 requires julia ≥ 1.12 (see JET v0.11.3 Project.toml), so it's incompatible with the lts (1.10) matrix entry under that forced-latest constraint. Non-dependabot PRs (like this one) don't trigger that flag, so the resolver is free to downgrade JET to 0.9/0.10 and resolve cleanly.

In other words, #75 fails for unrelated CI-config reasons, not because LogExpFunctions 1.0 is broken. The compat widening itself is safe.