Fix segfaults when assigning to list column

[tlapak]

Closes #4166
Closes #4667
Closes #4678
Closes #4729

The issue there was unrelated to the unlisting. A minimal example that also segfaults is

dt <- data.table(a=list(1:2, 3, 4))
dt[, a:=1:4]

Turns out when LHS is a list column and RHS isn't, no length checks were being made. I am slightly surprised by @MichaelChirico's result which was maybe even intended.

While testing my fix I noticed that subassigning a vector of correct length also segfaults:

dt <- data.table(a=list(1:2, 3, 4))
dt[1:2, a:=1:2]

I see three possible ways of handling this:

1. Error out with a coercion error
2. Treat RHS as length 1 and replicate
3. Coerce to list by wrapping elements of RHS in list()

I opted for 3. because why not. It seems the most intuitive to me besides 1. But happy to change this if people disagree. Required 'some' abuse of the BODY macro in memrecycle...

[codecov]

Codecov Report

Merging #4350 (5a811d1) into master (0cbc418) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #4350   +/-   ##
=======================================
  Coverage   99.47%   99.47%           
=======================================
  Files          75       75           
  Lines       14775    14785   +10     
=======================================
+ Hits        14697    14707   +10     
  Misses         78       78           

Impacted Files	Coverage Δ
src/assign.c	99.71% <100.00%> (+<0.01%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0cbc418...5a811d1. Read the comment docs.

[tlapak]

Uh, if I understand the Codecov report correctly, I need to add unit tests for raw and complex rhs and one that doesn't fit any of these types?

[jangorecki]

Thanks for PR. Yes, every new line of C has to be reached for code coverage to be happy. We can always workout code coverage later on. Much more important to pass the CI checks.

[tlapak]

Cool, thanks for the explanation. And I will try not to fat finger the close button again.

[jangorecki]

if I understand the Codecov report correctly, I need to add unit tests for raw and complex rhs and one that doesn't fit any of these types?

Yes, you understand correct. If you would push to Rdatatable org branch, it would be much easier for others to add those tests for you. If you cannot yet push branches to our org then @mattdowle should be able to invite you to members.

[tlapak]

I'm not a member so can't push to this repo.

Also, sorry for the delay. My time is a bit more limited than usual atm. I've now added the missing unit tests and found and fixed a bug in the process. I'm sure there's a lesson in there somewhere.