Don't check for a convention when not one isn't needed

[SeanTAllen]

trafficstars

When setting up filtering and sorting with UseFiltering and UseSorting, there are 3 different code paths that are traversed. Prior to this change, all 3 of them required that a convention be registered otherwise an exception would occur. However, for both filtering and sorting, only 1 of the code paths actually needs a convention. The other two do not use the convention at all.

This commit changes the filtering and sorting object field descriptor extensions to attempt to get a convention only within the block of code that requires a convention. By making this small change, the other two code paths that do not require a convention will no longer encounter an exception. Instead, filtering and sorting can be set up as expected.

This commit partially addresses #6983. We are able locally, with these changes applied, to use filtering and sorting from a type module by calling the versions of UseFiltering and UseSorting that take a type. Without the changes in this commit, we get the "no convention registered" error.

This change does not close #6983 as it is still possible to encounter the "no convention registered" exception and its suggested remedy of "Call AddFiltering() on the schema builder" doesn't work.

From what we can see, and have detailed in #6983, no conventions registered with the schema builder either directly or the defaults provided by AddFiltering and AddSorting are available at the time that a type module is run.

The result of this commit is to allow the usage UseFiltering and UseSorting with types created via a type module. The other option we are aware of is detailed by @jimitndiaye in https://github.com/ChilliCream/graphql-platform/discussions/6975. Jimit routes around the convention problem by not using a convention at all in his custom middleware. A close examination of UseFilteringInternal at https://github.com/ChilliCream/graphql-platform/discussions/6975#discussioncomment-8778742 will show this lack of convention usage as key to his fix. Instead of going the custom middleware route, we made these two small changes to Hot Chocolate. We prefer this approach.

[SeanTAllen]

My editor made a couple extraneous whitespace changes. Let me know if you would like me to revert them.

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 70.16%. Comparing base (ab3909e) to head (c762100). Report is 26 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #7019      +/-   ##
==========================================
- Coverage   72.85%   70.16%   -2.69%     
==========================================
  Files        2585     2585              
  Lines      129190   129491     +301     
==========================================
- Hits        94115    90857    -3258     
- Misses      35075    38634    +3559     

Flag	Coverage Δ
unittests	70.16% <100.00%> (-2.69%)	:arrow_down:

Flags with carried forward coverage won't be shown. Click here to find out more.

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[SeanTAllen]

@michaelstaib ping to get your attention on this per your comment https://github.com/ChilliCream/graphql-platform/issues/6983#issuecomment-2002461845

[SeanTAllen]

@PascalSenn in #6983, @michaelstaib said you should be pinged about this.

[PascalSenn]

@SeanTAllen, sorry, I didn't have this on my radar! @michaelstaib, this looks good to me. It's just moving the convention resolution into the branch that actually needs it.