fix: prevent escape fallback from swallowing malformed constructs and enable error test suite

[adi-IL]

Summary of Changes

1. Prevent Escape Fallback Swallowing Malformed Constructs:

   • In RegexParsers.java, the generic escaped character fallback string("\\").then(...) previously only excluded 'x' via isNot('x'). When specific escape constructs failed due to malformed or incomplete syntax (such as \N{} empty name, \N{abc unclosed name, \u123z invalid hex digits, \u incomplete unicode escape, \08 or \0 invalid octal escape), the parser fell through and consumed the leading character as a literal escaped character (Literal("N"), Literal("u"), Literal("0")).
   • Added .as("character name") to the \N{...} name parser.
   • Updated the generic escape fallback predicate to noneOf("0123456789xNu"), preventing digits, x, N, and u from falling through to literal single-character escapes when their specific syntax fails.

2. Enable and Expand Error Test Suite:

   • Renamed RegexParserErrorTests.java to RegexParserErrorTest.java (matching Maven Surefire's default **/*Test.java include pattern), allowing all 75 error tests to be automatically discovered and executed during standard mvn test runs.
   • Added tests in RegexParserErrorTest.java covering \u123z, \u, \08, and \0.
   • Updated ParsersRegexCheckTest.java in mug-errorprone to assert the compile-time diagnostic expecting <4 hex digits> for \u123z.

Validation

Executed full clean test suite across all 16 reactor modules:

• mvn clean test passed with 0 failures and 0 errors.
• dot-parse: 2,480 tests passed (including all 75 tests in RegexParserErrorTest).
• mug-errorprone: 929 tests passed.

[fluentfuture]

Thanks for submitting the fix!

AI identified a minor improvement. Please verify if it's applicable.

While this excludes digits, x, N, and u, it omits k, p, and P:

• Unclosed / Malformed Property (\p{foo / \P{foo): When \p{foo fails to parse as a valid property, the parser falls through to string("\\").then(one(noneOf("..."))) and successfully matches \p as a literal 'p', parsing \p{foo as Literal("p{foo") instead of throwing ParseException.
• Malformed Named Backreference (\k<foo): When \k<foo (missing closing >) fails word().between("\k<", ">"), it falls through and parses as Literal("k<foo").

Recommended Fix
Include k, p, P in the exclusion set.

The full list:

noneOf("0123456789xNuckpP")

[fluentfuture]

Let's use noneOf("0123456789xNuckpP")

[adi-IL]

Thanks for the review @fluentfuture! Added c, k, p, and P to the exclusion set (noneOf("0123456789xNuckpP")) and added test cases in RegexParserErrorTest covering \p{foo, \P{foo, \k<foo, and \c.

[fluentfuture]

AI suggested me to also change the one(isLetter, "category") to one("[a-zA-Z]").as("category") because JDK regex allows only ascii for the unbraced form.

If the use of isLetter is intentional, please speak up.

[adi-IL]

Oh yeah that makes total sense, definitely was not intentional. In JDK regex the unbraced single letter properties like \pL are strictly ASCII [a-zA-Z], and non-ASCII characters like \pα get rejected with an unknown property error.

Switching to one("[a-zA-Z]").as("category") is definitely the right call and fits dot-parse prefix pruning nicely too.

Thanks again for the quick review and merge.