Summary

When InstanceStart returns an error (e.g. container not found), both strategies now handle it properly instead of silently ignoring it.

Problem

InstanceInfoWithError.Error was set when InstanceStart failed, but neither strategy checked it:

• Dynamic: sessionStateToRenderOptionsInstanceState only used v.Instance, ignoring v.Error. The zero-value InstanceInfo{} was rendered as an empty loading screen that refreshed forever.
• Blocking: RequestReadySession only checked session.IsReady() and the top-level error from RequestSession. Individual instance errors were never inspected, causing needless retries until timeout.

Fix

• Dynamic (internal/api/start_dynamic.go): When v.Error is non-nil, render the instance with the requested name, unrecoverable status, and the provider error. This uses the existing error rendering support in all built-in themes ({{ if $instance.Error }}). The page continues to auto-refresh (200 OK), which is by design — the reverse proxy plugins use the X-Sablier-Status header for flow control, and the auto-refresh allows recovery if the user fixes the issue (e.g. creates the missing container).
• Blocking (pkg/sablier/session_request.go): Check session.InstanceErrors() after each RequestSession call and return early with an error (500) instead of retrying until timeout.
• Core (pkg/sablier/session.go): Add SessionState.InstanceErrors() method that collects and joins individual instance errors.

Before / After

http://localhost:10000/api/strategies/dynamic?names=foobar&session_duration=5m&theme=hacker-terminal

Before

After

Fixes #853