and implement castPtr with coerce, which gives
    12% less allocation in reverse-complem
    7.3% less allocation in fasta.
    Binary sizes fell 0.1%.
as reported and discussed in #9163.