CharDeduplication was not designed to deduplicate tokens with length 7+
which could lead to high memory consumption. With this change tokens of
all sizes can be deduplicated.

https://github.com/eclipse-jdt/eclipse.jdt.core/issues/1743

A benchmark implemented in CharDeduplicationTest.main(String[]) shows
the new deduplication is performed at similar speed (.21s instead of
.16s) but deduplicates much more tokens (99% instead of 36%).