Draft: restore BDF step size after tstop truncation

[dpastoor]

Summary

• Track the BDF step size before set_stop_time truncates a step to land exactly on tstop.
• When TstopReached is returned, rescale the BDF difference history back to the pre-truncation step size so the truncated stop interval does not become the next proposed step size.
• Add a focused regression test and a runnable reproducer.

Why draft

This preserves the documented TstopReached behavior that state.t == tstop, but it changes the meaning of state.h immediately after a stop: after TstopReached, state.h is the restored proposed next step size rather than the just-used truncated interval. That may be the right behavior for checkpoint/restart, but it is worth maintainer review before treating this as the final API contract.

This branch is independent of #1. It intentionally does not include the n_equal_steps checkpoint-state change.

Reproducer

Command:

cargo run -p diffsol --example bdf_tstop_reproducer

Output from this branch:

BDF tstop checkpoint/restart reproducer
ODE: dy/dt = -0.1 y, t in [0, 10], 100 output targets
rtol=0.000001, atol=0.00000001, h0=0.001

single sweep                                    55
cold restart at each target                     2100
checkpoint/restart + set_stop_time              320
checkpoint/restart + overshoot/interpolate      268

The set_stop_time row remains above the single sweep because this branch does not include #1. The key behavior tested here is that stop-time truncation no longer pins the checkpointed/proposed step size to the dense output spacing.

Tests

• cargo test -p diffsol bdf_tstop_restores_pre_truncation_step_size -- --nocapture
• cargo test -p diffsol test_tstop_bdf -- --nocapture
• cargo check -p diffsol --examples
• cargo run -p diffsol --example bdf_tstop_reproducer