BootstrapAgent::sled_reset doesn't fully reset the sled

The bootstrap agent exposes an API to reset the sled, and it notes its intent: https://github.com/oxidecomputer/omicron/blob/80b5f7f2f9253695f13832c62dface4116dbb2e8/sled-agent/src/bootstrap/agent.rs#L983-L984

However, it doesn't currently fully reset the state:

1. There's no way to "uninitialize" the LRTQ salt which gets set in request_sled_agent. (See #3818 - once request_sled_agent has been called once with a rack_id, all subsequent calls must provide the same rack_id).
2. (I think?) It does not unadvertise the sled prefix that is advertised in request_sled_agent.
3. (Unsure) Does it clean up any configuration it has sent to the switches (e.g., NAT entries for NTP/external DNS/instance zones)?

I chatted with @andrewjstone briefly and he suggested we may want to do something more extreme: remove anything written to the M.2s (bootstore / ledger), still remove persistent storage as reset_sled does today, then reboot the sled. That would take care of in-memory caches like LRTQ (unsure if there are more of these I've missed); I'm not sure about undoing switch configurations.

Start in sled-agent/src/bootstrap/agent.rs at the reset_sled implementation and request_sled_agent, then review issue #3818 for the LRTQ salt behavior. Trace the sled prefix advertisement and switch configuration paths, including persistent bootstore/ledger state. Done means the chosen reset design clears all relevant state and its behavior is covered by tests.