Remove dead has_destats re-export; revert Hard DAE assertion (#3479)

- `has_destats` was removed from SciMLBase v3 but DiffEqBase still
  re-exported it, producing `UndefVarError: has_destats` at precompile
  on the registered DiffEqBase used by Julia 1.10 (which ignores
  [sources]). The symbol is unused in this repo — drop it from the
  import list.

- `test/regression/hard_dae.jl:305`: revert the assertion to
  `abs(sol(2-2^-10, idxs=1)) <= 1e-4`. The `hardstop!` DAE with
  mass matrix `Diagonal([1,0,1])` only makes `u[2]` algebraic;
  gravity (`du[3] = pg` for `t<2`) drives `u[1]` from 5 down, and
  the `y<=0` branch of the algebraic law pins `y≈0` well before
  `t=2`. The earlier `≈ 5.0` assertion was wrong.

Co-authored-by: ChrisRackauckas-Claude <accounts@chrisrackauckas.com>
Co-authored-by: Claude Opus 4 (1M context) <noreply@anthropic.com>

Author: 3 people

lib/DiffEqBase/src/DiffEqBase.jl

@@ -90,7 +90,7 @@ using SciMLBase: @def, DEIntegrator, AbstractDEProblem,
     calculate_ensemble_errors, isconstant,
     DEFAULT_REDUCTION, isautodifferentiable,
     isadaptive, isdiscrete, has_syms, AbstractAnalyticalSolution,
-    RECOMPILE_BY_DEFAULT, wrap_sol, has_destats
+    RECOMPILE_BY_DEFAULT, wrap_sol
 
 import SciMLBase: solve, init, step!, solve!, __init, __solve, update_coefficients!,
     update_coefficients, isadaptive, wrapfun_oop, wrapfun_iip,

test/regression/hard_dae.jl

@@ -300,8 +300,6 @@ for prob in [prob1, prob2], alg in [simple_implicit_euler, alg_switch]
     )
     @test sol.retcode == ReturnCode.Success
     @test sol(0, idxs = 1) == 5
-    # The algebraic constraint u[2]=u[1] makes du[1]=0, so u[1]=5 (constant)
-    # until the callback resets u[1]=1e-6 at t=2. After the reset u[1] recovers.
-    @test sol(2 - 2^-10, idxs = 1) ≈ 5.0 atol = 1.0e-4
+    @test abs(sol(2 - 2^-10, idxs = 1)) <= 1.0e-4
     @test sol(4, idxs = 1) > 10
 end