EnC: Check member modifier updates in semantic analysis

[tmat]

Analyzing modifier changes in semantic analysis has multiple advantages:

1. We can detect the actual impact of change on the generated code. In cases when changing a modifier does not affect the generated code we don't need to issue a rude edit.
2. When analyzing syntax of a type or its member without semantic information we do not know whether to report a rude edit or not - the type may be reloadable, in which case most rude edits are not applicable.

The PR also fixes a few issues uncovered by the changes in the analyzer.

[tmat]

@davidwengier

[tmat]

@davidwengier Seems like we are missing tests that update attributes on non-type members (specifically events).

[davidwengier]

Seems like we are missing tests that update attributes on non-type members (specifically events).

Thanks. Logged #54430

[davidwengier]

The actual modifier update bits looks good. The analyzer changes are difficult to reason about without debugging through, but I trust the tests are doing the right thing, so I'm sure it's fine. Plus you're fairly trustworthy too 😛

[davidwengier]

I'm sure this is covered by some other case (and hopefully tests), but this looks like an issue. Whether there is an initializer, we still need to report rude edits for attributes etc., right?

[tmat]

We call DeferConstructorEdit for constructors that emit member initializers and declarations with initializers. This takes care of handling symbol changes (attributes, modifiers) for the latter. The former is handled in AddConstructorEdits where we actually issue edits for the constructor.