[NFC][SILGen]: remove dead code in SILGenExpr

[jamieQ]

Removes the following two bits of code in SILGenExpr that were either dead or superfluous:

1. Some pattern matching logic to dig out nested function conversions and explicitly apply a sendable conversion. Per the comments this was nominally done to avoid emitting multiple conversion thunks. However, I was unable to find a case that produced such thunks any longer, and additionally, the condition gating the logic always evaluated to false, so it's been removed.
2. Removed a superfluous conditional branch in RValueEmitter::visitConsumeExpr.

---

• Explanation: Removes some unnecessary code.
• Scope: Removes two sections of logic in SILGenExpr that were either dead or superfluous.
• Issues: N/A
• Risk: Minimal risk. The change should not alter existing behavior.
• Testing: Ran existing tests.
• Reviewers: @slavapestov

[jamieQ]

@swift-ci smoke test

[jamieQ]

@gottesmm this is probably very low priority, but I'd like your opinion on whether these deletions are the right fixes or if there are more appropriate changes that should be made. Thanks!

[jamieQ]

Want to confirm that deleting the condition here is the correct fix, and there's not something else we should be emitting in one of the branches. In the case below (~L7300-7312) the codegen looks more complex in the nontrivial branch.

[jamieQ]

I tried to figure out if this was still happening, but I couldn't find a case that reproduced this behavior. Maybe it was fixed in #86729?

[jamieQ]

This is the reason why this code is effectively dead – the conjunction contains both subExprType->getExtInfo().isSendable() and !subExprType->getExtInfo().isSendable() (presumably this was a typo and the second one was supposed to be subSubExprType).

Does removal seem like the most appropriate change here, or would fixing the logic error be better?

[slavapestov]

I'm fine with deleting dead code that is not covered by tests.