Type public results as TypedDicts; bump to 1.1.0

[gitRasheed]

Replaces the dict[str, Any] result aliases with real TypedDicts. Results stay plain dicts at runtime — no behavior change, verified against the committed R cache with no regeneration.

What users get

• RegResult, MRegResult, StackResult, BoostResult, VarResult, MebootResult, SeasonalityResult (plus the inner shapes: RegFitted, RegPoints, RegPredInt, DerivativeTable, RegEquation, RegXStar, MRegPredInt, SeasonalityTable) — exported from nns so pipelines can annotate.
• Autocomplete on R-style keys ("Fitted.xy", "Point.est") and mypy-caught typos, in user code and inside the library: nns_stack/nns_boost's consumption of nns_reg results is now actually type-checked.
• Conditional shapes are finally written down: NotRequired marks the point_only/features_only/h=0 partial forms; docstrings state when equation, x.star, pred.int, and conf.int.* appear.

Design decisions

• nns_reg is overloaded on multivariate_call: the internal True contract returns bare RegPoints; the public form returns RegResult | MRegResult (2-D x dispatches to nns_m_reg, matching R).
• Functional TypedDict syntax for dotted R key names, same as the existing part.py precedent; class syntax elsewhere.
• Two shapes stay dynamic on purpose: nns_arma's percent-labelled pred-int keys ("Lower 2.5% pred.int") are runtime-computed R names, and m-reg fit tables key columns V1..Vn — both remain dict[str, NDArray] with the format documented.
• Typing surfaced facts now recorded in the types: Fitted.xy gains conf.int.* columns when confidence_interval is set; boost's feature.frequency is float64 (R numeric); meboot's degenerate inputs return partial dicts.

Release setup

Bumps version to 1.1.0 in pyproject/__init__/README (check_version_consistency passes) per the batched-release plan for the next PyPI push.

Verification

• Full suite, NNS_R_CACHE_ONLY=1 (existing cache, no regeneration): 2202 passed; the 3 test_practical_examples failures are the known stale-local-extension issue, green in CI.
• mypy --strict and ruff clean; runtime smoke confirms results are unchanged plain dicts.

Note: until #77 (numpy<2.5 pin) merges, this PR's CI rolls the numpy-2.5.1 runner lottery — a cache-miss failure on 3.12/3.13 would be that, not this change.