[FIX] qc_report consuming all db connections + queue jobs timing out

[DESm1th]

After the move to the new dynamic QC pages tons of QC jobs were submitted to generate metrics that were (for some reason) missing from the QC folders. This large number of jobs consumed all available database connections. This was because SQLAlchemy opens a transaction whenever database reads happen and doesn't close them until the next commit/rollback. dm_qc_report was reading some values from the database prior to generating metrics, which left connections open 'idle in transaction' for 30+ minutes while the metrics were created.

This PR:

• Adds a decorator that will ensure a transaction ends after a function it wraps exits (8e248ce)
• Refactors dm_qc_report a bit to ensure all transactions are closed before metric generation happens (20a3c5f)
• Fixes a bug where one DTI file missing its bvec/bval file could crash the whole script (a0d7604)
• Increases the memory requested for queue jobs, as it turns out that newer studies have data large enough to require more than 1G of memory per job (88c2bfb)