ENH: Accept 0 and 1 as flags, negative positions and the horizon alias of align

[bashtage]

Branch a90-val2-bool-positions-alias, 2 commits on a90-fix-main-tests.
Merge #953 first. It is independent of #955 (they merge cleanly). This PR
covers release decisions 66, 71 and 75.

• Flags accept 0 and 1 (decision 66). "Accept bool, np.bool_ ... and 1/0
  (int, np.integer types). Anything else is not bool enough."
  • check_bool accepts bool, np.bool_ and Python or NumPy integers equal to
    0 or 1, and returns a Python bool.
  • Another integer raises ValueError. Floats (including 1.0), strings, arrays
    and None (where it is not the documented default) raise TypeError. The
    message is now "must be True, False, 0 or 1".
  • This applies to the 42 flags of the package, among them rescale,
    low_memory (both still accept None), disp (which also keeps "off" and
    "final") and center.
  • arch.test(exit=..., append=...) now checks its flags too. Before, exit=2
    ran the whole test suite.
• Negative positions count from the end (decision 71). "(b), which is a
  superset of (a). Also document this in docstrings and make sure there is a
  release note."
  • A negative integer position p means nobs + p, as in Python; positions
    outside [-nobs, nobs] raise ValueError. Dates are unchanged.
  • first_obs and last_obs of fit and fix, and cutoff_to_index, act as
    slice bounds: last_obs=-1 drops the last observation and first_obs=-100
    uses the last 100.
  • start of forecast (model, results and volatility process),
    hedgehog_plot and find_index is an observation index: start=-1 is the
    last observation.
  • Every docstring states the rule.
• align="horizon" (decision 75). "Document it and make it an official
  alias."
  • It is documented in every forecast with align and in
    ARCHModelForecast.
  • The model forecast, which used to reject it, now accepts it through a
    shared check.
  • A ForecastAlignment alias is added to arch._typing.

The contributing guide and the release notes are updated: one entry for the
flags, and entries for negative positions and the alias in New features.

Tests:

• arch/tests/univariate/test_positions_and_alias.py (46 tests) compares each
  negative position with the equivalent positive one, with Python slicing, and
  with a hand-computed ARCH(1) forecast. 37 of them fail on the base.
• The flag and position rows of the validation tables are updated (471 changed
  rows fail on the base).

Fast suite 24,738 passed. The only failures are the 7 covariance dtype tests that
also fail on main; a separate PR fixes them. The golden battery is identical.
numpydoc lint 0; style clean.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.02%. Comparing base (a93ff73) to head (72acc6b).

Additional details and impacted files

@@           Coverage Diff            @@
##             main     #956    +/-   ##
========================================
  Coverage   99.01%   99.02%            
========================================
  Files         134      135     +1     
  Lines       29563    29859   +296     
  Branches     2290     2303    +13     
========================================
+ Hits        29271    29567   +296     
  Misses        239      239            
  Partials       53       53            

Flag	Coverage Δ
adder	98.96% <100.00%> (+0.01%)	⬆️
subtractor	98.96% <100.00%> (+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.