Use of git-annex import --clean-duplicates can cause data loss, where git-annex deletes content that it doesn't actually have a copy of (i.e. there is no duplicate).
### What steps will reproduce the problem?
I've written a quick'n'dirty test script which goes through a bunch of combinations and tests --clean-duplicates. Given an 'origin' repo containing 'a' and 'b' content and a clone of it ('import') which doesn't contain 'a' and 'b' content.
g-a import --clean-duplicates ~/tmp/importme (containing a, b and c) into 'import' after:
Origin is set to trusted in import, b is dropped from within origin:
b is deleted from importme even though no annexes have copies (reasonable, as origin is set to trusted and import thinks it has the content).
Origin is set to semitrusted in import, b is dropped within origin:
b is deleted from importme even though no annexes have copies (this is most likely one to bite people).
Origin is set to untrusted in import, b is dropped within origin:
b is deleted from importme even though no annexes have copies and git-annex has been explicitly told to not trust information about origin :( This is really surprising behaviour!
### What version of git-annex are you using? On what operating system?
* 5.20150409
* Arch Linux (git-annex-bin)
### Please provide any additional information below.
I can provide the script if it is wanted (coded in Perl, couple of non-core dependencies).
