GitLab

lava-outerr: Avoid any chance of buffering on Python side

This fixes an issue when output on feedback connections of FVP (aka
FastModels, for use with which this utility was written in the first
place) appeared only after the process terminated. The fix:

1. Runs Python interpreter with -u (ubuffered) switch.
2. Instantiates subprocess.Popen with bufsize=0 param.
3. Reads from process in small chunks, to avoid blocking
for long time.

Perhaps only one of these was needed to fix the issue at hand, but be
future-proof and apply all.

Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>