Setting `maxTokens` in `harness.json` does nothing on non-mantle models

[Tommypop2]

Description

The maxTokens setting in harness.json isn't properly parsed into AgentRenderConfig, so it never is passed to Strands. Therefore, setting this property has no effect.
Tested with Claude Sonnet 5, where the model would run out of tokens on relatively simple tasks (with thinking) through using up tokens.

Steps to Reproduce

1. Set maxTokens to a value
2. Run agentcore dev
3. Observe the "Max output tokens" entry in "Harness Settings"

Expected Behavior

Observe that the "Max output tokens" entry in "Harness Settings" is populated with the correct value

Actual Behavior

Observe that the "Max output tokens" entry in "Harness Settings" remained empty

CLI Version

0.23.0

Operating System

Linux

Additional Context

No response