Monitor_nD event lists are somehow (still) not written in "atomic" fashion on GPU

[willend]

Most event lists created using openacc end up empty.

As e.g. "normal" 2D histogram function, the issue likely occurs in the calculation of the "index" into the (currently) 1-dimensional Buffer array.

[willend]

It turned out that the cure was to do an "atomic capture" to define a local TRACE copy of the "particle count", to be used everywhere in TRACE...

In a CPU setting where only one thread is active at a time we can of course make do with the usual, global variable on the structure.

[willend]

PR with fix is incoming.

[willend]

New observation - atomic / mostly correct particle data block written in list model, but the latest related edits in monitor_nd-lib.c lead to a couple of side effects:

1. On CPU, there is always a first bogus event present in the file 0 0 0 0 0 0

2. On GPU, the particle "id" is not incremented correctly (often resides in col no. 4)