Restrict character property fuzzy matching to "pattern whitespace"

I wasn't aware of this Unicode property when
initially implementing this. It's a more restricted
set of whitespace that Unicode reccommends for
parsing patterns. It's the same set of whitespace
used for extended syntax.

UAX44-LM3 itself doesn't appear to specify the
exact set of whitespace to match against, but this
is no more restrictive than the engines I'm aware
of.

Author: hamishknight

Sources/_RegexParser/Regex/Parse/CharacterPropertyClassification.swift

@@ -18,7 +18,7 @@ extension Source {
     // This follows the rules provided by UAX44-LM3, including trying to drop an
     // "is" prefix, which isn't required by UTS#18 RL1.2, but is nice for
     // consistency with other engines and the Unicode.Scalar.Properties names.
-    let str = str.filter { !$0.isWhitespace && $0 != "_" && $0 != "-" }
+    let str = str.filter { !$0.isPatternWhitespace && $0 != "_" && $0 != "-" }
                  .lowercased()
     if let m = match(str) {
       return m

Tests/RegexTests/ParseTests.swift

@@ -2061,6 +2061,16 @@ extension RegexTests {
       """, changeMatchingOptions(matchingOptions(adding: .extended))
     )
 
+    parseWithDelimitersTest(#"""
+      #/
+      \p{
+        gc
+         =
+        digit
+      }
+      /#
+      """#, prop(.generalCategory(.decimalNumber)))
+
     // MARK: Delimiter skipping: Make sure we can skip over the ending delimiter
     // if it's clear that it's part of the regex syntax.
 
@@ -2486,6 +2496,10 @@ extension RegexTests {
     diagnosticTest(#"\p{aaa\p{b}}"#, .unknownProperty(key: nil, value: "aaa"))
     diagnosticTest(#"[[:{:]]"#, .unknownProperty(key: nil, value: "{"))
 
+    // We only filter pattern whitespace, which doesn't include things like
+    // non-breaking spaces.
+    diagnosticTest(#"\p{L\#u{A0}l}"#, .unknownProperty(key: nil, value: "L\u{A0}l"))
+
     // MARK: Matching options
 
     diagnosticTest("(?-y{g})", .cannotRemoveTextSegmentOptions)