feat: add a new internal analyze_files tool with supporting infrastructure for file attachment handling

[cristian-groza]

Key Changes

• New analyze_files tool: Internal tool for analyzing file attachments with configurable analysis types
• Job attachments module: Extracted job attachment handling logic into dedicated job_attachments.py module with improved organization
• Job attachment wrapper: New wrapper class for managing file attachments in agent workflows
• Internal tool factory: Factory pattern implementation for creating and managing internal tools
• Dependency updates: Updated uipath and jsonschema-pydantic-converter package versions

Technical Highlights

JSON Schema Converter

The converter solves a critical challenge with dynamically generated Pydantic models by:

• Creating a shared pseudo-module (jsonschema_pydantic_converter._dynamic) in sys.modules that acts as a namespace for all dynamically created types
• Recursively collecting and registering all nested BaseModel types with proper module attribution
• Ensuring get_type_hints() can resolve forward references when models are used with external libraries like LangGraph

Why this was needed: Dynamically generated Pydantic models need proper module references for type introspection to work correctly. Without this, frameworks that rely on typing.get_type_hints() (like LangGraph) fail to resolve forward references in nested models, breaking validation and serialization.

[andreitava-uip]

this is ok for now, but we will have to revise this at some point.

1. It doesn't provide a nice way to compose functionalities (like file handling & static args)
2. It risks other code relying on isinstance(AnalyzeFileTool). In my design, systems would only search of Mixins

[andreitava-uip]

nit: this is ok for now but we will likely change our internal state management in the near future

[andreitava-uip]

what happens if some of the paths overlap?
ie: $.attachments[*] vs $attachments[0]

In that case we will have duplicate matches in the result. Is that a problem?
If we assume we will get disjoint paths (which is the case for the paths you extract earlier), we could at least make it clear in the docstring.

[cristian-groza]

If the paths overlap the function will return duplicates. Currently this is not a problem because the get_json_paths_by_type function is producing disjoint paths by design. I will update the docstring to make this behavior clear.

[andreitava-uip]

i'm really not a fan of this... but there is no way around this.

• Our InputModel is dynamically generated from json schema -> unavoidable
• We need to dynamically extend the AgentGraphState with the InputModel, otherwise we do not have the inputs in the state -> almost unavoidable
• AgentGraphState is then inspected by langgraph internals with get_type_hints(), which will fail to resolve dynamic types unless we do this. We have no mechanism to pass a local namespace through.

[andreitava-uip]

Please squash commits before merging