Avoid normalizing SyntaxContext ids in test output #74800

PR #72121 affects the output of any test which includes debug-printed Spans in its stdout, since the printed SyntaxContext id will now depend on which foreign SyntaxContexts have been imported. However, the precise id assigned depends on what macros were expanded inside libcore/libstd, which depends on the platform we are targeting. As a result, including raw SyntaxContext in test output makes that test platform-dependent.

Currently, we strip out all SyntaxContext ids to make the test output platform-independent. However, it would be good to perform a more fine-grained normalization: e.g. distinguishing between root and non-root SyntaxContexts, which should hopefully not be platform dependent.