Requeuing MAgPIE fails if it happens while distributing input files

[orichters]

• if REMIND (and probably MAgPIE) is preempted with bad timing while unpacking, it does not get back on its feet
• Happened here with multiple REMIND automated model tests that run on standby nodes
• What happens is probably:
  • download_unpack() is executed which writes source_files.log
  • Before or while the unpacked files are distributed by download_distribute(), the run was preempted/requeued and so that was not finalized
  • when the run was restarted, the existence of source_files.log with the correct information made the run think everything is fine
  • it fails with something like cp: cannot stat 'config/gdx-files/indu_subsectors-buil_simple-tran_edge_esm-POP_pop_SSP5-GDP_gdp_SSP5-En_gdp_SSP5-Kap_debt_limit-Reg_62eff8f7.gdx': No such file or directory
• It might be sufficient to write the source_files.log only if the file distribution was successful, not the unpacking alone.

[orichters]

• in REMIND, we will fix that with an additional check that if not all input data files are present, input data is downloaded and distributed again
• I would leave it up to you, @tscheypidi, whether you still want to fix that in gms, whether you want to add a similar check to MAgPIE (assuming this problem exists there as well) or none of the above and simply close here.