Add CensoredGARCH: latent-variance correction for censored return series

[uzair1509]

Implements a GARCH(p, q) volatility process for return series subject to a known, time-varying censoring rule (e.g. an exchange price-limit band, such as PSX's daily circuit breaker). When a return is censored (the true shock exceeded a known threshold but only the threshold is observed), using the raw truncated squared residual in the ordinary GARCH recursion understates the true conditional variance.

CensoredGARCH corrects this by replacing the squared residual at a censored observation with its conditional expectation given censoring, under the model's own conditional variance, using the truncated-normal second-moment identity:

E[eps_t^2 | |eps_t| >= c_t] = sigma_t^2 * (1 + a_t * phi(a_t) / (1 - Phi(a_t)))
a_t = c_t / sigma_t

This is the correction described in Morgan & Trevor (1999) and applied to exchange price-limit censoring in follow-up work (Wei 2002, Hsieh & Yang 2009). A companion open-source package (mtgarch on PyPI) applies this same estimator to Pakistan Stock Exchange data; this PR ports the estimator into arch as a proper VolatilityProcess subclass so it can be used directly with arch_model's full estimation, inference, and forecasting machinery instead of a standalone implementation.

Changes

• New arch/univariate/censored_garch.py: CensoredGARCH class and its recursion (censored_garch_recursion), following the GARCH / FixedVariance classes as templates. Implements all required abstract methods (bounds, constraints, compute_variance, starting_values, parameter_names, simulate) plus one-step-ahead analytic forecasting.
• Recursion acceleration: optional numba JIT path (~100x faster than pure Python), with a pure-Python fallback used automatically when numba is not installed, consistent with how arch treats numba elsewhere.
• arch/univariate/__init__.py: export CensoredGARCH.
• arch/tests/univariate/test_censored_garch.py: unit tests covering parameter/constraint setup, input validation, reduction to plain GARCH when nothing is censored, a direct check of the recursion against the truncated-normal identity, edge-case branches of the recursion, and an end-to-end ConstantMean + CensoredGARCH fit/forecast smoke test.
• Docs: added to the volatility-processes autosummary and changelog.

Known limitations

• Only horizon-1 analytic forecasting is supported; multi-step forecasts would be a natural follow-up.
• No Cython backend (numba and pure Python only).

[codecov]

Codecov Report

❌ Patch coverage is 99.74490% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 98.98%. Comparing base (80d3688) to head (c51b360).

Files with missing lines	Patch %	Lines
arch/tests/univariate/test_censored_garch.py	99.57%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff            @@
##             main     #875    +/-   ##
========================================
  Coverage   98.97%   98.98%            
========================================
  Files         136      138     +2     
  Lines       30028    30420   +392     
  Branches     2327     2355    +28     
========================================
+ Hits        29719    30110   +391     
  Misses        255      255            
- Partials       54       55     +1     

Flag	Coverage Δ
adder	98.92% <99.74%> (+0.01%)	⬆️
subtractor	98.92% <99.74%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[uzair1509]

Added tests for the uncovered branches in censored_garch.py. Coverage is now 100% (line and branch), with and without numba. Ready for review.

[uzair1509]

Merged the latest main and fixed the conflict in volatility.rst. Also fixed the lint issues (flake8, ruff, black, isort, numpydoc) and added tests for the uncovered branches in censored_garch.py (now 100% coverage). Ready for review.