Fix cold start psiphon/cache interaction in smart-dialer

[ohnorobo]

User issue while using smart dialer + psiphon

Going to background and resuming, when the android app is still running, and doing another check goes straight to Psiphon strategy and tests it successfully. I did find one usecase where it doesn't work exactly as expected. When ending up on a Psiphon fallback, storing that to outline strategy cache, and then doing a cold start of the app, it looks like Psiphon isn't quite ready at that point, it gets canceled and the whole DNS list starts to check. Then Psiphon wakes up, does it's test, and then does it again once all the DNS strategies fail. Sending a log with this cold start.

log: https://docs.google.com/document/d/1D4ZKq0cKcHT0jQnM5NqT9EWGy3paMe5s9Nx-YSFloJw/edit?tab=t.0

This eventually still starts working, but it requires re-running the entire search instead of using the cache.

The initial psiphon failure message is

Failed to start dialer: Psiphon: {PropagationChannelId: [REDACTED]} newPsiphonDialer failed: failed to start psiphon dialer: context canceled

Possibly an issue where the initial psiphon start context has been preserved from a previous startup but is obviously already cancelled.

This issue is after already applying the fix from https://github.com/Jigsaw-Code/outline-sdk/pull/523

[ohnorobo]

Event sequence:

• This initial findFallback call is failing after getting the old strategy from cache
• We retry the proxyless strategies here which fail
• We make another findFallback call that succeeds

The interior failure inside psiphon.go:Start here is just "context canceled", we don't have more log info on the exact context. There is both a startCtx and a dialerCtx there, but the dialerCtx should be freshly started from the background context on each call.

The ctx used in the first and second findFallback call is also the same.