Filter ANSI escape sequences for terminal colors as a new term type

[eldipa]

Describe the feature you'd like
byexample uses pyte to interpret all the ANSI escape sequences coming from the interpreter. But pyte is not cheap and introduces some artifacts in the output like truncated lines.

Some interpreters however doesn't use the full set of possible escape sequences. In the happier case, they would use only the escape sequences to add some coloring to their output.

These sequences don't change the final layout so filter them will yield the same result except colorless. This is what pyte does and it could be implemented directly in byexample with a simple regex.

To be backward compatible the new filter-the-colors mode could be added as a new terminal type

byexample/byexample/runner.py

Lines 546 to 551 in a084bd6

	if options['term'] == 'dumb':
	out = self._emulate_dumb_terminal(self._output_between_prompts)
	elif options['term'] == 'ansi':
	out = self._emulate_ansi_terminal(self._output_between_prompts)
	elif options['term'] == 'as-is':
	out = self._emulate_as_is_terminal(self._output_between_prompts)

, as an intermediate level between dump and ansi.

We could make the change non-backward compatible and add the filtering in dump as it is unclear how many people could be really testing ANSI escape sequences in their examples.

[eldipa]

The use of a filter/terminal emulator by default or extending +dumb is not a good idea.

First, processing the escape sequences is required only in a few cases. Do the filtering by default will slow down all the executions with a minimal of benefits.

Moreover, the terminal emulator may change slightly the way the output is obtained and the whole thing does not pay the price of breaking examples.

[eldipa]

A way to speed up the things and get no negative side effect is to use a specialized termscrapper screen/stream to mostly do nothing except ignore the escape/control sequences.
Internally the screen will just store the feed data in a list without any concept of a 2D screen working indeed as a byte linear filter.
Virtually this does not add any overhead and it should be safe to enable it by default.