Exponential decay when requests fail

[hbruch]

This PR addresses entur/lamassu#340. It introduces an UpdateStrategy which schedules new requests depending on the number of previously failed request attempts, with a max delay of one hour.

To verify this behaviour, a unit tests is added, which asserts that immediately after a failing request the feed should not update. Note that this is a rather implicit blackbox test. I considered extracting UpdateStrategy to a proper class instead of a private nested class to do a whitebox test, but refrained from this as GBFSFeedUpdater is a private inner class also.

Note further, that I came across two questions regarding the current implementation, which probably should be addressed in another PR (or need not, depending on the current reasoning).

[testower]

Thanks for your contribution @hbruch! Regarding your questions, I think they are relevant when design the v2 loader api. I would very much like to discuss this with you when I get there. The UpdateStrategy will probably be continued in v2, but then as a public api, so that the strategy can be tweaked by the library user if desired.