Auto merge of rust-lang#140485 - Jarcho:from_expansion_opt, r=<try>

Optimize the codegen for `Span::from_expansion`

See https://godbolt.org/z/bq65Y6bc4 for the difference. the new version is less than half the number of instructions.

Also tried fully writing the function by hand:
```rust
sp.ctxt_or_parent_or_marker != 0
        && (
            sp.len_with_tag_or_marker == BASE_LEN_INTERNED_MARKER
            || sp.len_with_tag_or_marker & PARENT_TAG == 0
        )
```

But that was no better than this PR's current use of `match_span_kind`.

Author: bors

compiler/rustc_span/src/span_encoding.rs

@@ -306,8 +306,16 @@ impl Span {
     /// Returns `true` if this span comes from any kind of macro, desugaring or inlining.
     #[inline]
     pub fn from_expansion(self) -> bool {
-        // If the span is fully inferred then ctxt > MAX_CTXT
-        self.inline_ctxt().map_or(true, |ctxt| !ctxt.is_root())
+        // Optimizes much better than `inline_ctxt` since it avoids comparing against `lo_or_index`.
+        let ctxt = match_span_kind! {
+            self,
+            InlineCtxt(span) => SyntaxContext::from_u16(span.ctxt),
+            InlineParent(_span) => SyntaxContext::root(),
+            PartiallyInterned(span) => SyntaxContext::from_u16(span.ctxt),
+            // Must be the value stored in `ctxt_or_parent_or_marker` to optimize properly.
+            Interned(_span) => SyntaxContext::from_u16(CTXT_INTERNED_MARKER),
+        };
+        !ctxt.is_root()
     }
 
     /// Returns `true` if this is a dummy span with any hygienic context.