PercentCodec decoder to disallow non-ASCII characters in input

[ok2c]

@arturobernalg Could you please double-check?

[arturobernalg]

I am afraid this implementation changes the existing behavior in two problematic ways.

CharsetDecoder#newDecoder() defaults to REPORT, while flushBin ignores the returned CoderResult. Unconsumed bytes are then retained by compact() and can be combined with a later percent-encoded sequence across a literal character. For example, decoding %C3x%A4 as UTF-8 currently produces \uFFFDx\uFFFD, whereas this change produces xä.

The encode fast path also bypasses charset encoding for the initial safe-character sequence. For example, encoding abc as UTF-16 currently produces %FE%FF%00a%00b%00c, whereas this change produces abc.

I think both cases need to be addressed before this can be merged.

void testDecodeDoesNotCombineEscapedBytesAcrossLiteralCharacters() {
    assertEquals(
            "\uFFFDx\uFFFD",
            PercentCodec.decode("%C3x%A4", StandardCharsets.UTF_8));
}

@Test
void testEncodeHonorsUtf16Charset() {
    assertEquals(
            "%FE%FF%00a%00b%00c",
            PercentCodec.encode("abc", StandardCharsets.UTF_16));
}```

[ok2c]

@arturobernalg Thank you so much for a thorough review. I will re-design the whole thing and will likely take a different, simpler approach.

[ok2c]

@arturobernalg I dropped all attempts at memory optimization and made the decode method treat non-ASCII characters in input as invalid / unrecognized.

Please do another pass.

[arturobernalg]

@ok2c LGTM