Handle recursive store for primitive types symbol/string

[wsmoses]

No description provided.

[github-actions]

Benchmark Results

	main	ffa3f27...	main/ffa3f27a7aa63c...
basics/overhead	4.34 ± 0.002 ns	4.34 ± 0.002 ns	1
time_to_load	1.1 ± 0.0019 s	1.09 ± 0.0028 s	1.01

Benchmark Plots

A plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR.
Go to "Actions"->"Benchmark a pull request"->[the most recent run]->"Artifacts" (at the bottom).

[vchuravy]

Is this the same as #2263

[wsmoses]

your fix might be more general, but also not completely solve the issue. Essentially from here we have this fallback case if we know the object is a jl_valuet and we just put a null pointer there. Here we hardcode adding string/symbol (but ensuring we actually do the null store), whereas yours has a more general 0 check but doesn't store (potentially still having the GC issue)

[vchuravy]

So we should likely do both? Aka it makes no sense to try and store into a datatype without any fields, but we should also create a shadow store for those objects.

[vchuravy]

Check #2263?

[wsmoses]

@vchuravy we should discuss this next week. going to merge to unblock flux test infra