In case we loose the race to update the reference, return GIT_EMODIFIED
to let the user distinguish it from other types of errors.

This changes the behavior of callbacks so that the callback error
code is not converted into GIT_EUSER and instead we propagate the
return value through to the caller.  Instead of using the
giterr_capture and giterr_restore functions, we now rely on all
functions to pass back the return value from a callback.

To avoid having a return value with no error message, the user
can call the public giterr_set_str or some such function to set
an error message.  There is a new helper 'giterr_set_callback'
that functions can invoke after making a callback which ensures
that some error message was set in case the callback did not set
one.

In places where the sign of the callback return value is
meaningful (e.g. positive to skip, negative to abort), only the
negative values are returned back to the caller, obviously, since
the other values allow for continuing the loop.

The hardest parts of this were in the checkout code where positive
return values were overloaded as meaningful values for checkout.
I fixed this by adding an output parameter to many of the internal
checkout functions and removing the overload.  This added some
code, but it is probably a better implementation.

There is some funkiness in the network code where user provided
callbacks could be returning a positive or a negative value and
we want to rely on that to cancel the loop.  There are still a
couple places where an user error might get turned into GIT_EUSER
there, I think, though none exercised by the tests.

This adds giterr_user_cancel to return GIT_EUSER and clear any
error message that is sitting around.  As a result of using that
in places, we need to be more thorough with capturing errors that
happen inside a callback when used internally.  To help with that,
this also adds giterr_capture and giterr_restore so that when we
internally use a foreach-type function that clears errors and
converts them to GIT_EUSER, it is easier to restore not just the
return value, but the actual error message text.

There are a number of cases where it is convenient to be able to
fetch and "claim" the current error string, clearing the error.
This is helpful when you need to call some code that may alter
the error and you want to restore it later on and/or report it via
some other mechanism.

The filter registry as implemented was too primitive to actually
work once multiple filters were coming into play.  This expands
the implementation of the registry to handle multiple prioritized
filters correctly.

Additionally, this adds an "attributes" field to a filter that
makes it really really easy to implement filters that are based
on one or more attribute values.  The lookup and even simple value
checking can all happen automatically without custom filter code.

Lastly, with the registry improvements, this fills out the filter
lifecycle callbacks, with initialize and shutdown callbacks that
will be called before the filter is first used and after it is
last invoked.  This allows for system-wide initialization and
cleanup by the filter.