File backend improvements

This merge request is intended to mitigate NFS issues with running iDQ in production by how we access files. The two main things this changes are:

1. Change file access pattern:
   • previous: write to final location -> copy to tmp file -> move to /dev/shm
   • now: write to temp directory -> copy to /dev/shm -> move to final location

The idea here is that the heavy write operation is done in a local drive. The copy (if we request this) is also to a local disk to /dev/shm. The only operation that interacts with NFS is a move, and that should be instantaneous on our end.

2. Remove the nested directory structure to store data products.
   • previous: train-gstlal_run5/START-12382/1238290518_1238292973/ANN1train-1238290518-2455.pkl
   • after: train-gstlal_run5/START-12382/ANN1train-1238290518-2455.pkl

The point of doing this is to avoid the massive number of directories created by idq-timeseries. The directory structure is a bit redundant, since the filename contains the same information as the directory 1238290518_1238292973. For timeseries, it would avoid creating a directory for every frame file, which also avoids some of NFS issues we've been having.

As a result, I had to make a few modifications to make this work:

• Add tmpdir property to DiskReporter, which takes in an optional tmpdir option from the INI file. If not set, looks for a TMPDIR environment variable. Else, defaults to /tmp.
• Modify DiskReporter.report() to make the file access pattern change.
• Modify DiskReporter.directory() to point at the new directory structure.
• Modify DiskReporter.glob() to be able to access files correctly.
• Add glob2path in names.py, used in DiskReporter.glob().

Also some house cleaning:

• Remove a duplicate report() in SegmentReporter
• Remove unused reporters JSONMetricReporter and DruidReporter

This has been tested in batch and streaming jobs, found no issues with operation or in creating reports.