[evlatests] modcomp-free: fast switching, flags, quack.

[Vivek Dhawan]

I got an hour on saturday and spent it switching between
3C84 = 0319+415 and 0313+413 just 1.5deg away.
X-band, continuum, 1.6sec integration.

The project name and the first few minutes of data that appeared
after the start of the file (according to the operator log) were
from another file (SYSSTART), but had no fringes. This may just
be a feature of the first observation after a changeover to
modcomp-free operation. It is a gotcha - if one loads data by
VLAOBS (project) name, none appears.

FILLM can load data with & without online flags applied, and
comparing the two datasets shows that:

1. With online flags enabled, the data are essentially perfect.

    Literally a handful ~4 visibilities on ea24, 2 on va25 were
    low amplitude, cause unknown. A few dozen points on baselines
    to antenna 5 were very high, also cause unknown, I have not
    looked at whether the online flags should have caught these.

    (13 and 23 were weak/dead but that is another matter)

2. The online flags are cutting off a bit of good data, 0 or
    1 integrations at the start of each scan, and a bit more,
    usually 2 or 3 but occasionally upto 5 integrations at the
    end of each scan. I believe from previous emails that this
    tolerance is band-dependent, and might bother someone doing
    very short scans at Q band, is it worth testing?

3. I tried QUACK (not every opcode) on the data loaded without
    any flags enabled. It appears to work and removes the end
    effects but more of a shakeout is clearly needed.

4. I offset the weaker source about 30" away from field center,
    but the sidelobes from it at the center are still mJy-ish.
    Therefore, though I see no sign of a fake source at field
    center, this particular test is best done on a blank field.
    An earlier short test showed the fake source gone. A deep
    integration is needed next.

Vivek.