per @grebe on #974 and #1006:
I think it is useful to treat count as being offset by 1 so count=0 means 1 symbol, count=1 means 2 symbols, etc. Disallowing count=0 is wasteful when count is narrow, e.g. in the case of count being 1 bit it will always be longer. If you encode count with an offset, it should probably be named something like offset_count or count_minus_one to avoid confusion.
per @grebe on #974 and #1006: