constantly assure that the terminals are in operational mode.

They should not drop out of operational, but if we are slow, they
do.

we still have to artificially tell that this is a float
value, although that's actually in the SDO.

we used to initialize the sync manager to what was
in the eeprom, which is big enough normally, but not always.

Now we read out the pdo assignment first, and set the sync
manager size accordingly.

for most operations, pre-operational is good enough a state.
By thinko we went to safe operational, which is not needed.

f is already used for float, so use x, not great but the first
free letter.

the Leybold TurboVac does not like to be put into operational unless
it is already getting PDO data. This actually seems like a good idea.

See merge request !10

we were already able to use huge values in assignments, now
we can also use them in operations.

this can easier be done earlier

as Python makes a difference here, follow its logic and call
floor division as such.

This is the basics to starting to support Leybold turbo pumps, which have some peculiar quirks.

Also general improvements.

See merge request !9

the anybus client does not like our index datagram with zero
length data. While we are already at it, we also do proper padding
according to the ethercat standard. Hopefully this way we are
compatible with many more devices.

somehow the anybus thing wants to return a EoE frame when
asked for a CoE frame. We can just ignore it, then it works
fine.

some terminals do not support reading 8 bytes of the eeprom at
a time, we have to limit to 4 bytes at a time

this involves restructuring quite a bit, probably with compatibility
problems.

this is not actually ethercat specific

the format characters already include the information whether they
are signed or long, no need to add this information again.

this means we know what values were set in ebpf. It is a bit
magic how often this will get called.

all kinds of stuff updated:

- EEPROMs can now be written, especially the serial number
- tests should work again
- ebpf devices can now act cyclic like slow devices

See merge request !7

even when using fast ethercat, some information may be perfectly
valid slow as well. This way we have the best of both worlds.

and use it instead of our own random number generator for the
missed packages simulator

this tests now the actual initialization of the terminals.

I removed the ebpf tests for now, as I am about to change them anyways.

don't use RuntimeError, define an EtherCatError instead

those didn't work anymore since we have mmap