Add fast path for ascii chars to isFirstCodepointUppercase

Whenever we know that the Rope is either using ASCII or something that
is a superset of ASCII (such as UTF-8), we can make a faster check
without needing to extract codepoints and checking for a bigger range
of uppercase characters.

I've used the instructions on
<https://github.com/oracle/truffleruby/blob/master/doc/contributor/profiling.md#require-profiling>
(in combination with a native binary -- `jt build --env native` +
 `jt -u native ruby ...`)
to benchmark the impact of this change and the numbers were too close to
the margin of error to tell:

* `master` (without the fix to #2079): avg 57.7ms, median 57ms, p90 64ms
* this commit: avg 57.8ms, median 58ms, p90 63ms
* without fast path (previous commit): avg 56.7ms, median 56ms, p90 63ms

...but since this fast path is both very simple and local to the method,
I think it's worth keeping anyway.

Author: ivoanjo

src/main/java/org/truffleruby/parser/lexer/RubyLexer.java

@@ -3668,8 +3668,15 @@ protected boolean isSpaceArg(int c, boolean spaceSeen) {
     /** Encoding-aware (including multi-byte encodings) check of first codepoint of a given rope, usually to determine
      * if it is a constant */
     private boolean isFirstCodepointUppercase(Rope rope) {
-        byte[] ropeBytes = rope.getBytes();
-        int firstCharacter = rope.encoding.mbcToCode(ropeBytes, 0, ropeBytes.length);
-        return rope.encoding.isUpper(firstCharacter);
+        Encoding ropeEncoding = rope.encoding;
+        int firstByte = rope.get(0) & 0xFF;
+
+        if (ropeEncoding.isAsciiCompatible() && isASCII(firstByte)) {
+            return StringSupport.isAsciiUppercase((byte) firstByte);
+        } else {
+            byte[] ropeBytes = rope.getBytes();
+            int firstCharacter = ropeEncoding.mbcToCode(ropeBytes, 0, ropeBytes.length);
+            return ropeEncoding.isUpper(firstCharacter);
+        }
     }
 }