Be a little more accurate/safe with bounding boxes

[svank]

PR summary

I remembered last night that Astropy has pixel centers at integer coordinates, so the image actually starts at -0.5, not 0, and so when we find a bounding box for mosaic reprojection (and also in the celestial intermediary code), the points we position along the edges of the input image should be along (x,y) = -0.5.

This PR puts them at -1, just for good measure. (The other side was already at shape[0] or shape[1], so it was already just a smidge past the actual edge which is at shape[0] - 0.5.)

I don't know why I'd originally set the points at (x,y) = 1 instead of (x,y) = 0. I'd count that as a bug, though it apparently wasn't affecting the test case I did yesterday.

Test plan

This just makes the bounding box bigger, so it shouldn't affect anything.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 87.03%. Comparing base (872f76b) to head (1d7b735).
Report is 4 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #327   +/-   ##
=======================================
  Coverage   87.03%   87.03%           
=======================================
  Files          56       56           
  Lines        3734     3734           
=======================================
  Hits         3250     3250           
  Misses        484      484           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.