Handle null values in joined summary columns

[infisble]

Fixes #2656.

When summaries with different column providers are joined, a column may not have a value for benchmarks originating from another summary. SummaryTable stored that null value and later dereferenced it while calculating the column width, which caused the reported NullReferenceException.

This change renders missing column values as NA and adds a regression test that joins two summaries where a custom column only applies to one of them.

Tests:

• dotnet test tests/BenchmarkDotNet.Tests/BenchmarkDotNet.Tests.csproj -c Release -f net10.0 --no-restore --filter FullyQualifiedName~JoinedSummaryHandlesMissingColumnValues (1 passed)
• dotnet test tests/BenchmarkDotNet.Tests/BenchmarkDotNet.Tests.csproj -c Release -f net10.0 --no-restore --no-build --filter FullyQualifiedName~BenchmarkDotNet.Tests.Reports.SummaryTableTests (23 passed)
• Full BenchmarkDotNet.Tests run: 1022 passed, 4 skipped, 2 failed. The two Linux CPU parser fixture failures reproduce unchanged on a clean upstream/master worktree.

[timcassell]

I thought usually the non-existing values are displayed as - rather than empty. It would be good to unify the behavior.

[infisble]

Good point. I updated the fallback to use - and adjusted the regression test in 6c3b992. The focused test passes, along with all 23 SummaryTableTests.

[timcassell]

Actually looking at the code it seems the common display values for non-existant are NA or ?. - is used for 0 values in metric columns. I think NA fits better here.

[infisble]

Thanks for double-checking. I switched the fallback to NA and updated the regression test in 626fef8. The focused test passes, along with all 23 SummaryTableTests.