Reject non-finite randomized interval ranges

[OskarEichler]

Reject configuration whose individually finite max_interval and rand_factor multiply to a non-finite randomized upper bound. Otherwise rand receives an infinite range only when retry execution reaches backoff generation.

Validation is shared by Config and ExponentialBackoff and leaves ordinary finite ranges unchanged.

Verification: current suite 210 examples, dual-Ruby completion model 426 checks, six-file syntax, RBS and full lint pass. Two of 104 focused randomized-range cases fail before; all pass after.

[kamui]

Thanks!

1. Please add a regression spec here too!
2. I think floating point rounding could make these not equivalent. With max_interval = Float("0x1.0d79435e50d79p+1023") and rand_factor = 0.9, validation returns finite, but runtime max_interval + rand_factor * max_interval produces Infinity; rand then raises Errno::EDOM.
3. A finite runtime range can be rejected. The spec says the change “leaves ordinary finite ranges unchanged.” With max_interval = Float("0x1.89d89d89d89d8p+1023") and rand_factor = 0.3, validation produces Infinity, while the runtime expression produces finite.