PROOF-OF-CONCEPT: allow deleting files that are in use

[dscho]

This should not be merged as-is, but is intended to act as a discussion-starter.

Applying the insights of https://github.com/LloydLabs/delete-self-poc, this commit implements support for deleting files that are in-use, even if that is theoretically impossible on Windows.

The ramifications are somewhat non-trivial, though:

• Many Windows applications are totally unhappy when the contract "files that are in use cannot be deleted" is broken.

• Even git.exe itself can be used to demonstrate the dire downsides of this here patch: when switching branches in git-sdk-64 where, say, mingw64/bin/libintl-8.dll needs to be updated (which is a dependency of mingw64/bin/git.exe), a checkout.workers value greater than 1 will let git.exe attempt to spawn git checkout--worker which will fail to load because of the missing dependency (which is missing because git.exe just deleted it).

This is a potential alternative to #1666, without the loss of performance of that Pull Request.

[dscho]

I tried using this a couple of times with git pull in the git-sdk-64 checkout I have over here, and ran into serious trouble:

• Since there are many files to update, I use checkout.workers=56. This means that 56 (!) git checkout--helper processes are spawned. However, they are spawned only after libintl-8.dll, a dependency of git.exe, is deleted, which means that the git checkout--helper processes all fail because of that missing DLL.
• When the in-use msys-2.0.dll was updated, everything started failing afterwards (I had a lot of sessions open, so that was kind of unpleasant).

Granted, not many users want to use Git to upgrade Git.

Having said that, I expect many similar problems should we merge this PR: It is such a long-standing contract that Windows programs can rely on files remaining present as long as they are in use.

I guess we could guard this behind a new config setting, but I am worried that this is a real foot gun.

[dscho]

Looks like there are more problems: https://github.com/git-for-windows/git/actions/runs/7038769040/job/19156545431?pr=4719#step:5:390 (and it's not only win test, win+VS test agrees):

failure: t5516.95 fetch exact SHA1
[...]
  ++ git fetch -v ../testrepo 9ad36e1e54b2130a20d55abb4f0f3ca8494ead3f:refs/heads/copy main:refs/heads/extra
  remote: error: object file ./objects/50/b820aea6d3503362343cdc0e699b760c700b2b is empty        
  remote: error: object file ./objects/50/b820aea6d3503362343cdc0e699b760c700b2b is empty        
  remote: fatal: bad tree object 50b820aea6d3503362343cdc0e699b760c700b2b        
  error: git upload-pack: git-pack-objects died with error.
  fatal: git upload-pack: aborting due to possible repository corruption on the remote side.
  remote: aborting due to possible repository corruption on the remote side.
  fatal: protocol error: bad pack header
  error: last command exited with $?=128
  ++ rm -rf child
  ++ exit 128
  ++ eval_ret=128
  ++ rm -rf testrepo
  ++ exit 128
  ++ eval_ret=128
  ++ :
  not ok 95 - fetch exact SHA1

failure: t5516.96 fetch exact SHA1 in protocol v2 
[...]
  ++ git -C child fetch -v ../testrepo 9ad36e1e54b2130a20d55abb4f0f3ca8494ead3f:refs/heads/copy
  remote: error: object file ./objects/50/b820aea6d3503362343cdc0e699b760c700b2b is empty        
  remote: error: object file ./objects/50/b820aea6d3503362343cdc0e699b760c700b2b is empty        
  remote: fatal: bad tree object 50b820aea6d3503362343cdc0e699b760c700b2b        
  error: git upload-pack: git-pack-objects died with error.
  fatal: git upload-pack: aborting due to possible repository corruption on the remote side.
  remote: aborting due to possible repository corruption on the remote side.
  fatal: protocol error: bad pack header
  error: last command exited with $?=128
  ++ rm -rf child
  ++ exit 128
  ++ eval_ret=128
  ++ rm -rf testrepo
  ++ exit 128
  ++ eval_ret=128
  ++ :
  not ok 96 - fetch exact SHA1 in protocol v2

[dscho]

I made up my mind: This change would cause more problems than it solves, so I'll abandon the idea.