Have the gateway continue to poll despite there being an error in experimental_updateSupergraphSdl

Hello 👋
We're utilizing the `experimental_updateSupergraphSdl` function to have the gateway pick up any changes to the supergraph schema. This comes in handy for us in a staging environment where the gateway can be in front of subgraphs/services that have different branches deployed to constantly.

We noticed that the polling stops when there's an error in `experimental_updateSupergraphSdl`.
This can happen when the supergraph schema doesn't exist yet (on the initial boot of the gateway) or if there's a bad write to the supergraph schema file. Here is the general idea of what we have:

```
experimental_updateSupergraphSdl: async () => {
const supergraph = new Supergraph()
const sdl = await supergraph.readSdl()

return {
id: crypto.createHash('md5').update(sdl as string).digest('hex'),
supergraphSdl: sdl
}
}
```

It'd be great if the polling can continue despite there being an error or a bad `supergraphSdl`.
This would be useful for when the app is "self-healing" and has a way to compose a valid supergraph schema after the failed attempt.

Thank you!

Start at the implementation of experimental_updateSupergraphSdl and the gateway's polling loop. Reproduce the case where reading the supergraph fails or returns an invalid schema, then verify that polling continues and a later valid schema is picked up. Done means transient errors no longer stop future polling.

Written by the indexing model from the issue text.