Tighten execution_context typing from Any to proper forward reference

[mlieberman85]

Context

#189 added execution_context fields to two dataclasses with loose typing:

packages/darnit/src/darnit/sieve/handler_registry.py:104:

execution_context: Any | None = None

packages/darnit/src/darnit/sieve/models.py:55:

execution_context: Optional["ExecutionContext"] = None  # already uses forward ref

The CheckContext side uses a proper "ExecutionContext" forward reference (wrapped in TYPE_CHECKING), but HandlerContext uses Any | None. This means custom handler authors lose type-check assistance when calling handler_ctx.execution_context.get_or_run_tool(...).

Proposal

Change HandlerContext.execution_context to use a proper forward reference matching CheckContext:

if TYPE_CHECKING:
    from darnit.core.models import ExecutionContext

@dataclass
class HandlerContext:
    ...
    execution_context: "ExecutionContext | None" = None

Why the loose typing was chosen

Likely to avoid a circular import between sieve/handler_registry.py and core/models.py. TYPE_CHECKING + string annotation is the standard workaround and is already used in sieve/models.py.

Acceptance criteria

• HandlerContext.execution_context type is tightened to ExecutionContext | None
• No runtime circular import introduced
• mypy/pyright can infer .get_or_run_tool() on the attribute
• Existing tests still pass

References

• Introduced by: feat: implement shared ExecutionContext for concurrent sieve checks #189
• Related: Wire first heavy tool (e.g., OpenSSF Scorecard) to ExecutionContext.get_or_run_tool #194 (wiring first real consumer)