* hardlink behavior.

If I had a hardlinked file, I would expect both files to be updated 
at the same time. Currently, it will only update one of the files.

mkdir 1
touch 1/a
ln 1/a 1/b
cp -al 1 2
cowdancer

echo test > 2/b

will result in 2/b being different to 2/a.

This has implications, and I think this should be documented.

* count number of hardlinks

before searching for the file in i-node list, I can decide that 
if the hardlink count=1, it's definitely not hardlinked.
Would it be worth the change? benchmark it.

* Investigate fchmod/fchown
It is partially implemented, but difficult to really do.
open(, RDONLY) and fchmod/fchown is hard to do, because we're handling
inodes at that timing.

* cpbuilder implementation
A prototype exists, but I need a good test environment with network
connection, and a chunk of time allocated for it. A Codefest ?
