Fix (?) dependency tracking for the metavar span table

[petrochenkov]

metavar_spans is a piece of global data that lives in SessionGlobals, it is similar to hygiene data in the same SessionGlobals.

This table is populated during macro expansion and after that it doesn't change.
Data from the table can be read from any query that performs the span1.to(span2) operation.

This table is clearly a piece of untracked data.
We also cannot querify accesses to it by passing tcx to all span1.to(span2) operations.
I suspect the whole table should be hashed at some point after macro expansion so it becomes an initial input to the query system.

[petrochenkov]

This may be fixed in #134478.

[petrochenkov]

Yep, fixed by #134478.