Disable exponential backoff elapsed time limit

Prior to this change, our backoff policy inheritied the default total elapsed time limit of 15m.

Meaning, for example, if a transcode ran for 16m and then failed due to an upstream disruption (e.g., shellbox deployment), it would not be considered eligible for retry, failing immediately.

Instead, make the elapsed time limit configurable, but disabled by default by setting it to 0 [0]. This way, backoff is bounded only by the retry limit, and transcode duration does not "count against" the ability to be retried.

Also restructures tests a bit to better faciliate overrides of the default config therein.

[0] https://github.com/cenkalti/backoff/blob/720b78985a65c0452fd37bb155c7cac4157a7c45/exponential.go#L176

Bug: T385225