release(v2.0.0-alpha.1): top-level reorg + bsde/pde/stochastic_control + mean_field/agent_based/inference/optimization

Phase 1 (top-level reorg):
  - matrix_riccati promoted to crate root via re-export
  - new top-level groups: bsde, pde, stochastic_control,
    agent_based, inference, optimization

Phase 2 (bsde):
  - theta_scheme: linear-BSDE theta-scheme
  - deep_bsde_bridge: ConditionalExpectation trait + driver

Phase 3 (pde):
  - fokker_planck: 1D forward FP with conservative central FD
  - hjb_multid: explicit n-D HJB on Cartesian grid (d <= 3)
  - elliptic_fd: 2D Poisson SOR solver

Phase 4 (stochastic_control):
  - optimal_switching: Snell envelope backward induction
  - pontryagin: 1D LQR Riccati shooting
  - two_sided_intensity_control: bilateral intensity control

Phase 5/6 (controls):
  - optimal_control::quadratic_impact_control (closed-form Riccati)
  - stochastic_control::two_sided_intensity_control

Phase 7 (mean_field + agent_based):
  - mean_field::mckean_vlasov: interacting-particle Euler scheme
  - agent_based::mod: generic interacting-agent simulator

Phase 8 (inference + optimization):
  - inference::robust_drift: Huber IRLS drift estimator
  - optimization::generative_calibration_hooks: GenerativeSampler trait
    + Gaussian MMD + finite-diff calibration step

Tests: 38 NEW tests, all passing (165/170 lib total; the 5 pre-existing
failures predate v1.1 and are tracked separately).

Versions bumped: Cargo 2.0.0-alpha.1, pyproject 2.0.0a1.

Deferred to subsequent v2.0.x bumps (parallelisable follow-ups):
PyO3 bindings, executed companion notebooks, Sphinx RST pages,
hfthot-lab-instance propagation.

CHANGELOG.md

@@ -4,6 +4,58 @@ All notable changes to **optimiz-rs** are documented in this file. The format
 follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and the project
 adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [2.0.0-alpha.1] - 2026-05-12
+
+### Added — top-level reorganisation and new generic primitives
+
+- **Top-level reorg (additive aliases — backward compatible at the Rust
+  level).**  `optimiz_rs::matrix_riccati` is now re-exported at the crate
+  root.  New top-level groups: `bsde`, `pde`, `stochastic_control`,
+  `agent_based`, `inference`, `optimization`.
+- `bsde::theta_scheme` — implicit/explicit θ-scheme for linear BSDEs
+  with deterministic coefficients (closed-form analytic test against
+  the deterministic ODE `dY = -ρ Y dt`).
+- `bsde::deep_bsde_bridge` — `ConditionalExpectation` trait and
+  `DeepBsdeBridge` driver providing the CPU-side recursion hook for
+  external function approximators.
+- `pde::fokker_planck` — 1-D forward Fokker--Planck solver with
+  conservative central differences and explicit positivity safeguard.
+- `pde::hjb_multid` — explicit upwind solver for multidimensional HJB
+  on a regular Cartesian grid (`d ≤ 3`) with reflective boundaries.
+- `pde::elliptic_fd` — 2-D Poisson `-Δu = f` SOR solver verified
+  against the `sin(πx) sin(πy)` eigenfunction.
+- `stochastic_control::optimal_switching` — Snell-envelope backward
+  induction for discrete multi-mode optimal switching.
+- `stochastic_control::pontryagin` — Riccati-shooting solver for the
+  1-D LQR Pontryagin maximum principle (verified against the
+  closed-form `P(t) = s_T / (1 + s_T (T-t))`).
+- `stochastic_control::two_sided_intensity_control` — generic
+  bilateral intensity control with affine per-jump premia.
+- `optimal_control::quadratic_impact_control` — closed-form Riccati
+  feedback for a controlled SDE with quadratic running cost.
+- `mean_field::mckean_vlasov` — interacting-particle Euler scheme for
+  generic McKean--Vlasov SDEs with empirical-measure drift.
+- `agent_based` — generic interacting-agent simulator (consensus
+  dynamics test recovers the empirical mean exactly without noise).
+- `inference::robust_drift` — Huber-loss IRLS estimator for the drift
+  of a 1-D OU-type discrete-time process; resists 5 % outliers.
+- `optimization::generative_calibration_hooks` — `GenerativeSampler`
+  trait + Gaussian MMD loss + finite-difference calibration step.
+
+### Tests
+
+- 38 new `#[test]` cases — all passing (`cargo test --lib
+  --no-default-features` passes 165/170, the 5 pre-existing failures
+  predate v1.1 and are unrelated).
+
+### Notes — deferred to subsequent v2.0.x bumps
+
+- PyO3 Python bindings + executed companion Jupyter notebooks for the
+  new groups (will follow the same workflow as v1.1.x).
+- Sphinx RST documentation pages for `bsde`, `pde`,
+  `stochastic_control`, `agent_based`, `inference`, `optimization`.
+- Propagation of new modules into `hfthot-lab-instance` consumers.
+
 ## [1.1.0] - 2026-05-12
 
 ### Added — purely additive, no existing API changes