Increase test tolerances

[svillemot]

Needed for passing the testsuite on Debian unstable/sid armhf.

See https://buildd.debian.org/status/fetch.php?pkg=octave-control&arch=armhf&ver=4.2.1-2&stamp=1769258436&raw=0

[ttl-octave]

Thanks for the patch. Maybe the accuracy of the steady state solution computed as "ground truth" is not sufficiently high. Would you mind to test increasing the loop count in the %! # asymptotic filter equations to 30 without increasing the tolerance?

[svillemot]

Actually I realize that the problem is not linked to the ARM architecture.

The failure comes from the fact that the test is random. I could generate one failure on x86-64 using the following:

for i=1:100; test kalman; endfor

(in one of the runs, I got an error slightly above the tolerance of 2e-5, but smaller than the 3e-5 that I suggested in this PR).

I let you decide how you want to fix this. For the time being, the patch in this MR has been applied to the Debian package.

[ttl-octave]

Good catch! I needed a loop over 1000 test to get a failed one. The input vector for the process noise is randomly chosen in one of the tests. I am merging the pull request and will increase the accuracy of the steady state computation afterwards.