An actually buggy fallthrough!  "c" is clobbered.

Most consumers don't care about delays of 0, and nh-recorder is dead format
these days, so fixing this is not of high priority.

The right is but a placeholder.

Might be better to just use git these days...

It's harmful to exclude GPL=2.

Recording TELNET is a wee bit obsolete, and even logging other TCP
connections is of little use as everything networked is encrypted
these days.

Silences a bunch of warnings.

Instead of using 256 color mode for everything, with a hacky value of 16
for "default", and varying them at output -- tag the color with the mode.

Also, use textual description of attributes for the testsuite.

Colors will be tagged with the appropriate mode (mono/16/256/RGB) on bits
24..25 of the relevant attribute half (fg, bg).

They do nothing, but went out of bounds checking for CJK damage.

We'd need checks all around for CJK, and a terminal of width 1 is of no
real use anyway.

As ".cast-v1" is not a recognized read extension, this also exercises
autodetection.

This makes them work even if the file extension is not .cast (plus ev.
compression) -- unlike .ttyrec, .nh or .dm2, there seems to be no real
consistency.

As a downside, files that were supposed to be asciicasts but are not,
will get blasted via "live".

All because of ungetc()'s deficiency.