[BUG] multiprovider: a failed Init leaves the event channel open forever and never shuts down providers that did initialize

## Observed behavior

When `InitWithContext` fails, it returns early — before `close(p.outboundEvents)` and before `shutdownFunc` is assigned:

https://github.com/open-feature/go-sdk/blob/6b0824c6c7b9b69e40f26aedcc4406ad4d3a5cec/openfeature/multi/multiprovider.go#L403-L430

`ShutdownWithContext` then short-circuits on `!p.initialized`, so there is no recovery path:

https://github.com/open-feature/go-sdk/blob/6b0824c6c7b9b69e40f26aedcc4406ad4d3a5cec/openfeature/multi/multiprovider.go#L569-L576

Observed:

```
Init error = Provider bad: context deadline exceeded ; Status() = "ERROR"
shutdownFunc nil after failed Init = true
consumer goroutine on EventChannel() blocks forever after failed Init
```

Two leaks follow. Any consumer ranging over `EventChannel()` blocks permanently, since the channel is never closed. And inner providers that *did* initialize successfully never receive `Shutdown`, so their connections and goroutines leak.

## Expected Behavior

A failed `InitWithContext` should leave the provider in a state where cleanup is possible:

- close `outboundEvents` (or make `ShutdownWithContext` handle the partially-initialized state rather than short-circuiting), and
- shut down the inner providers that did initialize, per [1.1.2.3](https://openfeature.dev/specification/sections/flag-evaluation#requirement-1123) and [§2.5](https://openfeature.dev/specification/sections/providers#25-shutdown).

## Steps to reproduce

```go
// one provider's InitWithContext returns an error; the other succeeds
mp, _ := multi.NewProvider(multi.StrategyFirstMatch,
multi.WithProvider("bad", &failingInitProvider{}),
multi.WithProvider("good", &okProvider{}))

err := mp.InitWithContext(ctx, of.EvaluationContext{}) // non-nil
go func() { for range mp.EventChannel() {} }() // never returns
mp.ShutdownWithContext(ctx) // no-ops; "good" never shut down
```

Start in openfeature/multi/multiprovider.go at InitWithContext and ShutdownWithContext, then inspect EventChannel and the initialization path for partial success. Reproduce the failing-provider and successful-provider scenario from the issue. Done means a failed initialization permits cleanup, closes EventChannel, and shuts down providers that initialized successfully.

Written by the indexing model from the issue text.