Provide a method for admins to clear .well-known cache

This issue has been migrated from [#5139](https://github.com/matrix-org/synapse/issues/5139).

---

### Request
I'd like a way to force Synapse to retry the SRV records and .well-known/ files of a particular federation partner, perhaps a command line script I could run.

### Motivation
Recently, the lust.works Synapse server was misconfigured to report 8448 as the preferred federation port, rather than 443, breaking federation. I (matrix.nora.codes) attempted to create a room there during that time. Now that it's fixed, my Synapse still tries to connect at 8448, though its .well-known/ points (correctly) to 443.