gs: add {_strip/unstrip}_protocol

[dtrifiro]

fixes resolution of relative paths (iterative/dvc#7638)

fixes #38

[codecov-commenter]

Codecov Report

Merging #41 (9c92c56) into main (8b44ac4) will increase coverage by 0.77%.
The diff coverage is 50.00%.

@@            Coverage Diff             @@
##             main      #41      +/-   ##
==========================================
+ Coverage   47.00%   47.77%   +0.77%     
==========================================
  Files          46       46              
  Lines        2517     2545      +28     
  Branches      307      307              
==========================================
+ Hits         1183     1216      +33     
+ Misses       1311     1306       -5     
  Partials       23       23              

Impacted Files	Coverage Δ
src/dvc_objects/fs/implementations/gs.py	60.86% <50.00%> (-3.84%)	⬇️
tests/fs/test_system.py	100.00% <0.00%> (ø)
src/dvc_objects/fs/system.py	83.82% <0.00%> (+11.76%)	⬆️

---

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 8b44ac4...9c92c56. Read the comment docs.

[efiop]

@dtrifiro Could you please explain the issue and the solution? There is no info on it in the issue, so I'm trying to make sure I understand it correctly

[dtrifiro]

Since we were not stripping protocol for gs, the from_info in the snippet below started with gs://

dvc-objects/src/dvc_objects/fs/base.py

Lines 475 to 480 in feecfba

	to_infos = [
	localfs.path.join(
	to_info, *self.path.relparts(info, from_info)
	)
	for info in from_infos
	]

this resulted in relparts looking like this ('..', '..', '..', 'import-url-failing-test', 'directory', 'bar') when calling dvc import-url gs://import-url-failing-test/directory ( with directory containing bar) thus ending up writing the importer directory in a a directory starting with ../../../

[efiop]

@dtrifiro I guess it raises another question: should fs.path be able to handle full urls? This is an aspect of fsspec path handling that I'm not sure about. On the one hand it is handy, but feels like path normalisation would be nice to default to. So maybe fs.path should need normalize the path, similar to how _strip_protocol does. This is just a thought though. Your fix is perfectly valid for now.

[efiop]

Might want to check other filesystems too. Might have to come back to this later.