Update MutableSpan and OutputSpan

[meg-gupta]

• Annotate subscript and count of MutableSpan and OutputSpan with @_semantics. Span's subscript was already annotated with @_semantics and doesn't need a similar change.
• Use _assumeNonNegative for OutputSpan's count

[meg-gupta]

Based on #85476

[meg-gupta]

@swift-ci test

[meg-gupta]

@swift-ci

[atrick]

I'll just trust that the fixed_storage @_semantics are correct. They make sense to me but aren't documented.

If @glessard thinks it's worth adding new entry points in the stdlib to work around optimization barriers, then I won't object.

My primary concern is that we're avoiding fixing optimizer support for mark_dependence. The mark_dep should already be converted to [nonescaping]. Soon, it seems like we will also care about optimization in the presence of non-Copyable and borrowed values.

[meg-gupta]

@atrick I can document the semantics as a follow on. They have been in use for Span and InlineArray.

My primary concern is that we're avoiding fixing optimizer support for mark_dependence. The mark_dep should already be converted to [nonescaping]

Agree mark_dependence [nonescaping] is not handled well. But it didn't seem relevant for the tests whose bounds check optimizations were blocked in SIL due to the lack of semantics.

@glessard I would like to merge the semantics and _assumeNonNegative changes at the very least. BitwiseCopyable specializations are useful too but it's fine to not merge.

[glessard]

I agree with the semantics and _assumeNonNegative changes. The stdlib team's consensus is to prefer not to have the BitwiseCopyable specializations (and in fact we would like to remove them from Span.) How big is the gain from them? How distant is the prospect of them being no better than the optimizer?

[glessard]

@swift-ci please test

[stephentyrone]

I would expect that these have to be @inline(always), if we're going to have them.

[meg-gupta]

I removed the BitwiseCopyable specializations from this PR.

[meg-gupta]

@swift-ci test

[glessard]

Thanks @meg-gupta