Keep ad-hoc intersections when narrowing with a tuple of types

[mohitduhan19]

Fixes #22122

Problem

class A: ...
class B(A): ...
class C: ...

def f(x: A) -> None:
    if isinstance(x, (B, C)):
        reveal_type(x)  # before: "B"

B overlaps with A, so conditional_types() returns B | C. The ad-hoc intersection fallback in conditional_types_with_intersection() only runs when the whole result is uninhabited, so it never runs here. When the result is later narrowed against the declared type A, the C part is dropped and x becomes just B. A common subclass of A and C (e.g. class D(A, C)) still passes the check at runtime, so the narrowing is unsound, and mypyc-compiled code raises TypeError: B object expected; got D.

A single isinstance(x, C) already narrows correctly to <subclass of "A" and "C">.

Fix

When there are several type ranges and the overall result is inhabited, the yes-type is now built per range. A range that overlaps with the current type keeps its existing result. A range that is disjoint from it (an Instance on both sides) adds the same ad-hoc intersection that a single isinstance check would produce.

If no range is disjoint, the result is unchanged, so single-type checks and fully overlapping tuples behave exactly as before. isinstance(x, (B, C)) now narrows to B | <subclass of "A" and "C">.

Tests

Added testIsInstanceAdHocIntersectionTuplePartialOverlap (the issue's example, reversed tuple order, a fully overlapping tuple that stays unchanged, and a union declared type) and testIsInstanceAdHocIntersectionTuplePartialOverlapFinal (a @final disjoint class adds nothing).

Updated testIsInstanceWithStarExpressionAndVariable: with the test fixtures, int and tuple can have a common subclass, so the result now includes <subclass of "builtins.int" and "tuple[Any, ...]">. This matches what isinstance(var, tuple) alone already gives with the same fixtures. With real typeshed, int and tuple can't share a subclass, so users won't see this.

Locally, testcheck, testfinegrained, testdeps, testpythoneval and mypyc test_irbuild pass, and the self-check, black and ruff are clean.

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

spark (https://github.com/apache/spark)
+ python/pyspark/sql/session.py:1236: error: Argument "names" to "_inferSchema" of "SparkSession" has incompatible type "<subclass of "pyspark.sql.types.DataType" and "tuple[Any, ...]"> | list[str] | None"; expected "list[str] | None"  [arg-type]
+ python/pyspark/sql/session.py:1278: error: Argument "names" to "_inferSchemaFromList" of "SparkSession" has incompatible type "<subclass of "pyspark.sql.types.DataType" and "tuple[Any, ...]"> | list[str] | None"; expected "list[str] | None"  [arg-type]

meson (https://github.com/mesonbuild/meson)
+ mesonbuild/optinterpreter.py:127:24: error: Incompatible return value type (got "int | float", expected "TYPE_elementary | HoldableObject | MesonInterpreterObject | Sequence[TYPE_var] | dict[str, TYPE_var]")  [return-value]
+ mesonbuild/ast/interpreter.py:692:24: error: Incompatible return value type (got "int | float", expected "str | int | bool | HoldableObject | MesonInterpreterObject | <6 more items>")  [return-value]

apprise (https://github.com/caronc/apprise)
- apprise/apprise_attachment.py:186: error: Item "AttachBase" of "str | AttachBase | list[str | AttachBase | AppriseAttachment]" has no attribute "__iter__" (not iterable)  [union-attr]
+ apprise/apprise_attachment.py:186: error: Item "AttachBase" of "str | AttachBase | <subclass of "tuple[Any, ...]" and "apprise.apprise_attachment.AppriseAttachment"> | list[str | AttachBase | AppriseAttachment]" has no attribute "__iter__" (not iterable)  [union-attr]

scipy-stubs (https://github.com/scipy/scipy-stubs)
- tests/misc/test_submodules_empty.pyi:3: note: See https://mypy.readthedocs.io/en/stable/running_mypy.html#missing-imports