GaspPredict

It appears to happen more with longer strings and it looks like it's always the first 32 bytes od the output that get corrupted It could be caused by race...

The weird randomness probably stems from garbage collector rather than threading. Adding RB_GC_GUARD to loop in decompress_buffered as per https://docs.ruby-lang.org/en/master/extension_rdoc.html#label-Appendix+E.+RB_GC_GUARD+to+protect+from+premature+GC seems to also fix the problem ```diff diff --git a/ext/zstdruby/zstdruby.c...