fix: make WebSocket disconnects recoverable and actionable

[GiribaldiTTV]

Summary

Improve Tarkov.dev WebSocket resilience so recoverable disconnects do not
surface as noisy failures while required-send failures remain actionable.

This branch is rebased onto the current master, including the merged PR #212
Settings and profile-bound API-key work.

Why

The client previously treated an unexpectedly closed WebSocket as an
unstructured failure. This could leave the connection lifecycle unclear,
produce noisy diagnostics, and make later sends vulnerable to stale socket
state.

Changelog

User-visible

• Recoverable connection interruptions remain quiet while the client restores
  the connection lazily.
• A required send that cannot be completed produces one actionable diagnostic.
• Connection incidents are deduplicated so repeated failures remain readable.
• Existing Settings, profile-bound API-key management, and startup UI behavior
  from the current master are preserved.

Internal

• Serialize WebSocket lifecycle operations and sends.
• Add bounded reconnect backoff with cancellation-safe shutdown.
• Handle fragmented, malformed, binary, and partial frames safely.
• Avoid replaying messages after a partial batch has been sent.
• Redact sensitive connection details from diagnostics.
• Keep the socket change limited to DiagnosticModels.cs, MainBlazorUI.cs,
  MessageLog.cs, and SocketClient.cs.

Validation

• Current-master baseline Release build completed with zero errors.
• Reconciled branch Release build completed with zero errors.
• Self-contained win-x64 single-file publish completed with zero errors.
• Reconciliation completed without conflict markers or deleted files.

Known limitations

• Recovery still depends on the Tarkov.dev service and the local network
  becoming reachable again.

[Razzmatazzz]

@GiribaldiTTV could you please resolve the merge conflicts? Thanks!