sem: replace addr calls with nkAddr

[zerbina]

Summary

In typed AST, address-of operations are now always represented by
nkAddr trees. This simplifies some compiler logic, makes processing
for typed macros easier, and fixes an effect tracking bug with addr.

Details

This is effectively a revert of
nim-lang/Nim#10814.
Not turning calls to mAddr into nkAddr was done to prevent the
unsafe address semantics from being lost, but those no longer exist.

Lowering the call into an nkAddr tree has the benefit that it
simplifies AST analysis and processing, as address-of operation can now
always be detected by PNode.kind == nkAddr. Old code for detecting
mAddr magic calls is removed.

The effect tracking in sempass2 had no special case for mAddr
calls, thus treating them as normal calls, which led to addr(x) being
treated as an indirect invocation of x, when x is of procedural
type. With the mAddr call now being lowered earlier, this is no
longer the case.

[saem]

/merge

[github-actions]

Merge requested by: @saem

Contents after the first section break of the PR description has been removed and preserved below:

---

Notes for Reviewers

• unblocks docs: build documentation for threading-related modules #1464

[zerbina]

/merge

The tslow_tables.nim failure is definitely spurious (the timeout probably needs to be increased at some point), and the tinternal_thread_cleanup.nim failure too looks like it is spurious (though I haven't seen the test fail in the past).

I'm giving merging another try.