Jira: LE-307

Summary

• Fixed field_order arrays in 136 component nodes across 32 starter project JSON files
• The field_order was out of sync with the actual inputs definitions in the Python source files, causing parameters to display in the wrong order in the UI
• Affected components include: Chat Input, Chat Output, Language Model, Agent, Prompt Template, Text Input, Tavily AI Search, Read File, Embedding Model, Message History, Parser, Astra DB, FAISS, MCP Connection, Split Text, API Request, Write File, and others
• Added a new parametrized test (test_template_field_order_matches_component) that validates field_order in all starter project JSONs against the actual component definitions, preventing future regressions

Test: test_template_field_order_matches_component

A new backend test that runs against all 33 starter project JSON files:

1. For each node with a metadata.module, imports the actual Python component class
2. Instantiates the component and calls _get_field_order() to get the canonical input order from the inputs list
3. Filters the component's field order to only the fields listed in the JSON's field_order (since JSON may be a subset)
4. Verifies the relative order matches — catches cases where fields are listed in the wrong sequence

What it catches:

• Fields listed in wrong order relative to the component's inputs definition
• Fields in field_order that don't exist in the component (they'd be filtered out, causing a mismatch)

What it doesn't flag:

• Missing fields from field_order (it's allowed to be a subset)
• Nodes without metadata.module (e.g. note nodes)