fix: decouple mocking from evals

[cristipufu]

Description

Decouples mocking from evaluation context.

• Separates the mocking infrastructure from the evaluation system, enabling mocking to be used independently.
• Mocking can now be configured without requiring a full evaluation item.

Ref: #1117

Development Package

• Use uipath pack --nolock to get the latest dev build from this PR (requires version range).
• Add this package as a dependency in your pyproject.toml:

[project]
dependencies = [
  # Exact version:
  "uipath==2.5.10.dev1011483997",

  # Any version from PR
  "uipath>=2.5.10.dev1011480000,<2.5.10.dev1011490000"
]

[[tool.uv.index]]
name = "testpypi"
url = "https://test.pypi.org/simple/"
publish-url = "https://test.pypi.org/legacy/"
explicit = true

[tool.uv.sources]
uipath = { index = "testpypi" }

[tool.uv]
override-dependencies = [
    "uipath>=2.5.10.dev1011480000,<2.5.10.dev1011490000",
]

[cristipufu]

@akshaylive why do we need this agentInfo?

[akshaylive]

The prompts were mostly copied over from URT (check L41). It does make sense for mocker to know about the agent's context -- especially the eval inputs. Maybe we need to create a separate class like this?:

class MockItem(BaseModel): # Equivalent to evaluation item
    inputs: Any
    name: str = Field(default="debug", ..)
    mocking_strategy: MockingStrategy

[cristipufu]

do we actually need this, or can we rely on mocker_context?

[akshaylive]

Yes. The reason is because the mocking strategy can be stateful. For example, a mockito strategy could say "when input is any, return "foo" for the first invocation and "bar" for the second invocation". Without keeping the reference to MockingStrategy in the context, this is hard.

[akshaylive]

Looks great overall. We do need to figure out the details regarding agentInfo and evaluation_criterias in a debug context.

[akshaylive]

It's strange for input mocker to be using expectation values. The prompts were reused so we didn't think much of this. Do you know what happens in prod during simulations -- is it {}?

[akshaylive]

Secondly, evaluation_criterias is a map of evaluator_id -> criterias. For URT/"legacy evaluation items", these are repeated during load, so this object will have a lot of repeated values.

My $0.02 is that we should completely get rid of these fields from input simulation but we can do that in a separate PR. @bai-uipath : could you follow up on this with the right POCs?

[akshaylive]

The prompts were mostly copied over from URT (check L41). It does make sense for mocker to know about the agent's context -- especially the eval inputs. Maybe we need to create a separate class like this?:

class MockItem(BaseModel): # Equivalent to evaluation item
    inputs: Any
    name: str = Field(default="debug", ..)
    mocking_strategy: MockingStrategy