fix off-by-one out-of-bounds write in EmulatedCharset.getBytes

[metsw24-max]

Noticed LatinCharset.getBytes writes bytes[i] while i runs from offset, but bytes is only count long. So OutputStreamWriter.write(buf, offset, count) with a non-zero offset on ISO-8859-1/Latin-1 writes past the output array and leaves bytes[0..offset) unwritten; the precondition check only validates offset/count against the input buffer. Index the destination with i - offset.

[niloc132]

Thanks - at a quick glance that looks correct, and confirmed that the test fails without the change (and passes with it).

The fork Google made also appears to need this fix: https://github.com/google/j2cl/blob/master/jre/java/javaemul/internal/EmulatedCharset.java

[metsw24-max]

Good point, the LatinCharset code is identical over there. I'll send the same one-line fix to j2cl.