fix(BudgetTerm.f90): ensure that the size of an array is >0 before deallocating

[emorway-usgs]

• Fix offered in BudgetTerm was discovered using models that included only a single SFT/SFE reach. Previously, GWF models with a single SFR reach would work. Only transport models would fail.
• 2 new autotests offered that each use only a single reach
  • The first autotest is for a GWT model (test_gwt_uztmvt2x1). This test ensures that in addition to a single reach, a single MVT connection between UZT and SFT works as expected
  • The 2nd autotest is for a GWE model (test_gwe_drycell_cnd3). This test checks that conduction between SFE and a dry host cell is working as expected in a single-reach model
• Extended an existing single-reach SFR autotest to include SFT. Prior to this fix, the added GWT model with a single SFT reach would fail because of the bug.
• Brief description of bug offered in release notes
• fprettify
• black

[christianlangevin]

I remember glancing at this a while back. I think I understand that you are intending to not write any flows for the single reach case and instead remove the flowjaface record entirely. I wonder if this is the best way to handle it. Perhaps it would be better to still include the header but just not write any records? I'm thinking that this may break other downstream processors, like zone budget. Do we have a similar problem for LAK/MAW/UZF? Can you remind me why this is a problem for SFT but not SFR (I think we discussed this once before).