[ty] Avoid overcounting shared memory usage

[ibraheemdev]

Summary

Use a global tracker to avoid double counting Arc instances.

[github-actions]

Diagnostic diff on typing conformance tests

No changes detected when running ty on typing conformance tests ✅

[github-actions]

mypy_primer results

No ecosystem changes detected ✅

Memory usage changes were detected when running on open source projects

flake8 (https://github.com/pycqa/flake8)
- TOTAL MEMORY USAGE: ~66MB
+ TOTAL MEMORY USAGE: ~69MB
-     memo fields = ~52MB
+     memo fields = ~54MB

trio (https://github.com/python-trio/trio)
- TOTAL MEMORY USAGE: ~159MB
+ TOTAL MEMORY USAGE: ~167MB
-     memo fields = ~119MB
+     memo fields = ~125MB

sphinx (https://github.com/sphinx-doc/sphinx)
-     memo fields = ~204MB
+     memo fields = ~214MB

prefect (https://github.com/PrefectHQ/prefect)
-     memo fields = ~424MB
+     memo fields = ~445MB

[github-actions]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

Formatter (stable)

✅ ecosystem check detected no format changes.

Formatter (preview)

✅ ecosystem check detected no format changes.

[MichaReiser]

Would it be worth considering to vendor our own get_size (in ruff memory usage). It's not very complicated code.

salsa-rs/salsa#944 would allow us to avoid the global lock.

[ibraheemdev]

It looks like untracked_arc_size was incorrect before in that it didn't account for the stack size of T, so the reported memory usage correctly goes up with this change.