Use a static context map with two buckets for UTF8 data.

Enabled for quality >= 4, and if there are no obvious
UTF8 violations detected.
For each block, we gather two separate histograms, one
for continuation bytes and one for ASCII or lead bytes.