Revert "[4.1] Stricter enforcement of the "large space" heuristic"

[xedin]

Reverts #13438

This PR caused some projects to fail the build. I will try to come up with reduced test-case and re-open the JIRA.

[xedin]

@swift-ci please test

[xedin]

/cc @CodaFi

[xedin]

@swift-ci please test source compatibility

[CodaFi]

I think I know what these projects might be... all the same the source compatibility break should be discussed because this is a case where the old Sema code demonstrably allows malformed Swift code through to SILGen.

[rudkx]

@CodaFi Malformed in the sense that we're not always diagnosing not-fully-covered switches, or in some other sense?

[CodaFi]

In the sense that SILGen currently inserts a pessimistic fatalError default branch that causes surprises to say the least. It’s not a full miscompile but it’s still undefined behavior caused by a bug in the compiler.

[rudkx]

I'm not seeing a fatalError inserted in the original test case from https://bugs.swift.org/browse/SR-6316, but I do see a crash if the code is optimized. It's not great behavior

Let's back this out of 4.1 for the moment and then discuss in https://bugs.swift.org/browse/SR-6652 whether we should adjust the heuristic for when it bails out.