refactor: expose macro evaluator in renderer

[benfdking]

See #4605 for context

[Copilot]

Pull Request Overview

Exposes the internal MacroEvaluator instance from the renderer so it can be accessed after rendering.

• Initializes and caches _macro_evaluator in the renderer constructor.
• Adds a read-only macro_evaluator property to return the last evaluated instance.
• Assigns the macro_evaluator during table resolution in _resolve_table.

Comments suppressed due to low confidence (3)

sqlmesh/core/renderer.py:76

• [nitpick] Consider expanding this docstring to clarify when this property is populated and what it returns if render hasn’t been called, to avoid confusion when consumers access it.

        """Returns the cached macro evaluator from the last render operation."""

sqlmesh/core/renderer.py:75

• Add a unit test that runs through the rendering process and asserts that macro_evaluator returns a valid MacroEvaluator instance, ensuring this new property is covered by tests.

    def macro_evaluator(self) -> t.Optional[MacroEvaluator]:

sqlmesh/core/renderer.py:72

• [nitpick] Consider resetting self._macro_evaluator at the start of each public render invocation (in addition to __init__) to prevent carrying over a stale evaluator when the same renderer instance is reused.

        self._macro_evaluator: t.Optional[MacroEvaluator] = None

[izeigerman]

This doesn't feel like a robust way of doing this. Can we instead have a method that constructs the MacroEvaluator instance? And then call it externally AND in _render?