syntax/utf8: avoid a spurious vector reallocation

This reworks `Utf8Sequences` logic in order to avoid allocating a
0-sized vector and immediately reallocating it for the initial element.
Directly create the populated vector instead.

I was looking at the memory usage patterns of [rolldown] through
heaptrack, and this spot showed up as a potentially-spurious temporary
allocation. The consumer side is [here][consumer side]. I do not have a
specific benchmark for this.

[rolldown]: https://github.com/rolldown/rolldown
[consumer side]: https://github.com/rolldown/rolldown/blob/ce36a195ed4e9ce7c446557cefff4750a2268e01/crates/rolldown/src/utils/extract_hash_pattern.rs#L12

regex-syntax/src/utf8.rs

@@ -302,9 +302,9 @@ impl Utf8Sequences {
     /// Create a new iterator over UTF-8 byte ranges for the scalar value range
     /// given.
     pub fn new(start: char, end: char) -> Self {
-        let mut it = Utf8Sequences { range_stack: vec![] };
-        it.push(u32::from(start), u32::from(end));
-        it
+        let range =
+            ScalarRange { start: u32::from(start), end: u32::from(end) };
+        Utf8Sequences { range_stack: vec![range] }
     }
 
     /// reset resets the scalar value range.