Graph optimization foundation with LangGraph/Graph adapter and Trace/OTEL/Sysmon backends

[doxav]

This PR adds an improved end-to-end approach for graph optimization in Trace, using LangGraph as the initial runtime target but aiming to be generalized to other graph structures.

Current scope:

• adds a graph adapter/module layer for graph-like runtimes
• supports graph optimization through three primary backends: trace, otel, and sysmon
• allows comparing multiple trace/observer views for the same execution
• keeps runtime outputs as normal Python objects while separating optimization-facing state
• integrates the graph/module path with existing optimization/search mechanisms such as PrioritySearch and ModuleCandidate (to be double checked by @chinganc )
• adds demos and tests to exercise the current design

This is still an early design base and some cleanup/refactoring is expected, but it should already be a useful basis for:

• reviewing how graph optimization fits in the current Trace stack
• discussing internal vs external tracing/optimization carriers
• evaluating whether this pattern could inform a broader evolution of the current bundle approach without modifying core primitives yet

Suggested entry point for review:

• docs/GraphOptimization.md

NO modification of core code => A clean-up/sync is required: some core files are modified just because branch is a little bit out of sync (e.g. utils.py, bundle.py...)

[chinganc]

TODO: integrates the graph/module path with existing optimization/search mechanisms such as PrioritySearch and ModuleCandidate

[chinganc]

LGTM