0.9.3 seemingly breaks panic documentation for `ReseedingRng`

[kayabaNerve]

0.9.2 documents the following:

Error handling

Although unlikely, reseeding the wrapped PRNG can fail. ReseedingRng will never panic but try to handle the error intelligently through some combination of retrying and delaying reseeding until later. If handling the source error fails ReseedingRng will continue generating data from the wrapped PRNG without reseeding.

Manually calling reseed() will not have this retry or delay logic, but reports the error.

1aeee9f#diff-e6e5ffc9c3a54105859ee7e76bb077b51313dff227e9bab169591e9bca95fc3d, as merged for 0.9.3, changed the failed reseed to a panic however.

[dhardy]

Thanks for the report. This is a doc issue, yes.

(I don't like changing the behaviour in a patch release like v0.9.3, but (a) the previous behaviour was unsound and (b) it is extremely unlikely that SysRng will fail during reseeding after successfully initializing a ThreadRng, hence probably we should ensure this is reported if it does ever happen.)

[kayabaNerve]

I'm fine if this is considered a doc issue. While I don't love a patch release changing behavior either, if you say it's unsound, that's enough for me to be quiet about it and defer to you 😅

This also did seem unlikely when I noticed it, but I didn't spend the time reviewing the exact APIs it was using and performing such an exhaustive map, so I can't comment there other than again deferring to you there, which I'm fine doing :)

[dhardy]

Fixed in #1763