Fix SIGSEGV when storage driver returns nil error (#3625)

[nonspecialist]

This addresses #3625 where tiller segfaults in uniqName -- the check for an existing name assumes that the storage driver's Get function always returns a valid error type, but for example the cfgmaps driver (default) often returns a nil

[nonspecialist]

A note on testing ... this is a hard one to test for; the only existing test that might trigger this is

helm/pkg/tiller/release_server_test.go

Line 357 in b4b693c

{"", "[a-z]+-[a-z]+", false, false},

and that would only trigger the issue if moniker chose the name happy-panda; with the current version of moniker having 243 animals and 390 descriptors, that's a 1 in 94,770 chance (0.001%) of triggering the bug during testing.

I could update this PR with a bigger rewrite that allows the use of something that satisfies the moniker interface in order to increase the test coverage, if that's considered worthwhile.

[nonspecialist]

PR updated to add a test which, based on the original code causes a segfault and confirms that the fix is valid

[bacongobbler]

hey! looks like the commit history is mangled. Mind taking a look into fixing that?

[nonspecialist]

Yeah, whoops ... sorry. Fixing that now

[bacongobbler]

LGTM. Thanks!