Compiler incorrectly uses UTF-8 length instead of ASCII for `ascii()` constant evaluation

[Gusarich]

The constant evaluator for the ascii() built-in function incorrectly interprets strings with \xHH escapes as UTF-8, causing it to count certain bytes (e.g., \xFF) as multiple bytes instead of a single byte. This misinterpretation leads to erroneous "ascii string is too long" compile-time errors for valid ASCII strings within the 32-byte limit.

Minimal Example:

const SeventeenFFs: String = "\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF"; // 17 bytes
const AsciiValue17: Int = ascii(SeventeenFFs); // Should compile successfully

contract TestAsciiLengthBug {
    val: Int;
    init() {
        self.val = AsciiValue17;
    }
}

Compiler Output:

Error: Cannot evaluate expression to a constant: ascii string is too long, expected up to 32 bytes, got 34

Expected Behavior:
Each \xFF escape sequence must be counted as exactly one byte (ASCII), resulting in a total length of 17 bytes, which is within the 32-byte limit.

Explanation:
The current implementation incorrectly calculates byte length using UTF-8 encoding, turning each non-ASCII codepoint (\xFF) into multiple bytes, thereby exceeding the allowed limit. The correct behavior is to treat each \xHH escape as exactly one byte, adhering to ASCII semantics.

---

LLM Fuzzing discovery (see #2490)