better way to handle scripts that export data from log directories

[doudou]

e.g. export streams to CSV files for matlab.

The issue right now is that the file names depend on the deployment choices - they change if someone changes the deployment. We should have a way to do it based on the stream name.

One way to handle this would be to finally normalize log generation to have a single file per stream. This would allow one to easily pick what stream he/she wants and needs and copy only that without complex changes. This is what the syskit-pocolog integration does when normalizing datasets, but we could move it to the logger itself instead. The downside is that it may be more CPU and I/O-intensive (more syscalls since we don't multiplex all streams into a single I/O anymore)

Another way would be to make pocolog (the command line tool) and pocolog (the ruby api) understand folders instead of single files, through e.g. a Pocolog::Directory API.