Reject non-ASCII chars in Base64.DecodeFromChars

[SimonCropp]

Bug

The OperationStatus overload of the Base64.DecodeFromChars polyfill copied the input into a UTF-8 buffer with (byte) source[i]. A non-ASCII char was narrowed to its low byte, which can be a valid base-64 character, so invalid input decoded successfully: "AQI\u0144" (U+0144 narrows to 'D') returned Done with the bytes of "AQID". The BCL returns InvalidData with 0 consumed and 0 written. (In a checked consumer build the same cast threw OverflowException.)

Fix

Chars above 0x7F are mapped to 0xFF, which is never a valid base-64 byte, so Base64.DecodeFromUtf8 reports InvalidData at the right position.

Test

Base64Tests.DecodeFromChars_OperationStatus_NonAscii covers a non-ASCII char at the end of the first quantum, at the start, and in the second quantum (asserting 4 consumed / 3 written). It fails with Done before the fix on net48. All Base64Tests pass on net48 and net8.0 (polyfill) and net11.0 (BCL).