Use single-affect VCC constructor in _track_callback

[ChrisRackauckas-Claude]

Please ignore until reviewed by @ChrisRackauckas.

Summary

• _track_callback(cb::VectorContinuousCallback, ...) was still calling the VCC constructor with a leftover nothing argument in the old affect_neg! slot. After the recent refactor that dropped affect_neg! from VCC, the constructor signature is (condition, affect!, len, ...) — the stray nothing made len::Nothing instead of Int, so the constructor threw a MethodError on every adjoint pass through any VCC, regardless of Julia version.
• Drop the stray nothing and forward initializealg, saved_clock_partitions, maybe_discontinuity, and initialize_save_discretes from the original callback so the tracked VCC round-trips all the user-supplied configuration.

Test plan

Reproduced the failure locally on Julia 1.12 via test/callbacks/vector_continuous_callbacks.jl. Before the fix: all 4 sub-testsets in this file error with MethodError: no method matching SciMLBase.VectorContinuousCallback(...). After the fix:

• MSE loss function bouncing-ball like → callback with linear affect — runs, with 4 numerical assertion failures (separate latent issue, see note below).
• Test condition function that depends on time only — passes.
• Structural simultaneous fire (algebraically tied conditions) — passes.
• Coincidental simultaneous fire (corner trap) — passes.

Note on remaining numerical failure

The callback with linear affect testset now reaches its assertions but they fail with ~7% mismatch on the u[3]/u[4] gradient components vs ForwardDiff. Both BacksolveAdjoint and GaussAdjoint agree with each other and disagree with ForwardDiff in exactly the same way, so this is a latent issue in the implicit ∇τ correction for VCC, not a regression from this PR. Master has been red on Callbacks2 for over a month and the MethodError was masking this. I deliberately kept this PR small and focused on the constructor fix per CLAUDE.md "small, focused PRs" guidance — the numerical issue deserves its own investigation and PR.

[ChrisRackauckas-Claude]

Update: also includes the test-point fix

Pushed commit 090adfe4 shifting u0[4] from 2.0 to 2.01 in test_vector_continuous_callback. This is the proper resolution to the numerical mismatch I noted in the original PR body — verified via central finite differences (see closed #1447 for the full check) that the failing assertions were at a jump discontinuity where ForwardDiff and adjoint legitimately compute different one-sided derivatives, not an algorithmic bug.

With both commits applied, all 11 tests in test/callbacks/vector_continuous_callbacks.jl pass on Julia 1.12.6 locally.

[ChrisRackauckas-Claude]

Correction: test-point fix moved to its own PR

Pushed a force-update reverting commit 090adfe4 from this PR. The test-point shift now lives in #1448 as a separate small PR. This PR is back to a single concern: the constructor MethodError fix in src/callback_tracking.jl.

Order doesn't matter — either PR can land first. With this one alone, 3/4 testsets in test/callbacks/vector_continuous_callbacks.jl pass and the 4th (callback with linear affect) fails numerically at a jump discontinuity (see closed #1447 for the central-FD writeup). #1448 then shifts the test parameter off the discontinuity so all 4 pass.