[NO-JIRA] Span/trace creations in the end of functions execution

[alexkuzmik]

Details

This PR changes the way track functionality works. Now CreateSpan or CreateTrace messages are generated and scheduled to be sent to the backend when the tracked function finishes working (before that CreateMessage was generated when the function began, UpdateMessage when it was ending).

Now UpdateSpan/UpdateTrace requests are sent to the backend only if you are using low-level api (which is still working as before).

The changes in the API and UX:

1. get_current_span() and get_current_trace() functions were removed. You can now use get_current_span_data(), get_current_trace_data() to get information accumulated for sending in the end of the current function. If you want to update it, you can use update_current_span(), update_current_trace(). All these functions are stored in opik.opik_context namespace.
2. Since CreateRequests are sent at the end of function executions, the trace will not be created on the backend side until all trace tree is executed. Which also means it won't be visible on the frontend side.

Implementation changes (high-level).

1. context_storage now operates not with span.Span, trace.Trace api objects, but with span.SpanData, trace.TraceData. Those objects have no methods that trigger any requests, they are simply dataclasses instances that store information about trace and span to be created. So, any part of the SDK that uses context_storage (or opik_context) has been adjusted accordingly.

Integrations:

I've updated the langchain integration, llama_index integration already. Other ones should be checked in case they are affected and updated accordingly.