[STJ SourceGen] Use Roslyn4.4 target when testing

[eiriktsarpalis]

Created this PR to try to isolate the test failures reported in #79828 (comment)

[ghost]

Tagging subscribers to this area: @dotnet/area-system-text-json, @gregsdennis
See info in area-owners.md if you want to be subscribed.

Issue Details

---

null

Author:	eiriktsarpalis
Assignees:	eiriktsarpalis
Labels:	area-System.Text.Json
Milestone:	-

[ViktorHofer]

I'm OK with not testing the roslyn4.0 source generator for now. We do the same for target frameworks and only test the very latest. If we notice regressions in the roslyn4.0 source generator though (i.e. based on customer reports), we should invest in testing it as well.

[eiriktsarpalis]

I neglected to put in the description that I created this PR in order to isolate some wasm test failures I've been seeing: #79828 (comment)

TL;DR running tests against Roslyn4.4 causes deterministic assertion failures when running release builds on wasm. I don't see how a compile time artifact such as the Roslyn version could trigger test failures at runtime, but I'll keep investigating.

[eiriktsarpalis]

Hi @dotnet/mono-team, it looks like the changes introduced in 5236bb8 are causing a number of tests to fail on wasm only. I've been able to narrow down the failures to a single unit test in fe3a627. The failure happens deterministically and only occurs on wasm targets. It appears to occur because the JsonDocument._utf8Json span is reported as empty, even though my changes in the second commit make this case impossible -- I suspect this could be caused by some kind of memory corruption.