fix(session): expand tool_call_id field limit & preserve tool function name in transcript

[nipunanr]

Summary

This PR resolves two critical issues in Flow session message persistence and LiteLLM/Gemini API integration:

1. Tool Call ID Truncation Error: Prevents CharacterLengthExceededError when models (such as Google Gemini, Antigravity, or deep-thinking models) generate long thought tool call IDs (>140 characters).
2. Missing Tool Call ID Exception in LiteLLM: Ensures role: "tool" messages include their corresponding function name both during execution loops and when reconstructing conversation transcripts from stored DB session messages.

---

Problem & Root Cause

1. tool_call_id Truncation (CharacterLengthExceededError)

• Symptom:

  Error: Flow Session Message, Row 16: 'Tool Call ID' (call_046343...) will get truncated, as max characters allowed is 140
  

• Root Cause: In Flow Session Message (flow_session_message.json), tool_call_id was defined as Data (mapping to VARCHAR(140)). When advanced LLMs emitted long thought/tool call IDs (>140 chars), Frappe raised a validation error during session.save(), causing the Flow run to fail.

2. LiteLLM / Gemini Tool Matching Error

• Symptom:

  litellm.APIConnectionError: Missing corresponding tool call for tool response message. Received - message={'role': 'tool', 'toolcallid': '...'}, last_message_with_tool_calls={...}
  

• Root Cause: LiteLLM's Gemini transformer requires every role: "tool" response message to specify its function name. If name is omitted, LiteLLM attempts to match tool_call_id against last_message_with_tool_calls. In multi-turn sessions or resumed runs with ID sanitization/mismatches, LiteLLM failed to resolve the name and threw an API error.

---

Proposed Changes

flow/flow/doctype/flow_session_message/flow_session_message.json & .py

• Changed tool_call_id fieldtype from Data to Small Text (text in MariaDB).
• Updated type annotations in flow_session_message.py to DF.SmallText | None.
• Benefit: Removes string length truncation checks while supporting IDs up to 65,535 characters.

flow/lib/agent.py

• Updated tool message construction in _loop() and _resume_loop() to explicitly attach "name": call.name to all role: "tool" messages.

flow/flow/doctype/flow_session/flow_session.py

• Enhanced _row_to_message() and transcript() to map preceding assistant tool_calls to their function names when reading stored transcript rows.
• Attached "name" to role: "tool" messages in transcript outputs.

---

How Has This Been Tested?

1. Database Schema Verification:
   • Ran bench migrate and verified that tabFlow Session Message.tool_call_id column successfully converted to text.
2. Unit & Integration Tests:
   • Executed bench run-tests --app flow. All 98 core agent unit tests passed cleanly (OK).

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[greptile-apps]

Confidence Score: 4/5

The PR is not yet safe to merge because subsequent and resumed prompts still omit persisted tool function names.

_build_prompt_messages() continues reconstructing rows without the ID-to-name map, leaving the previously reported LiteLLM/Gemini failure reachable.

Files Needing Attention: flow/flow/doctype/flow_session/flow_session.py

_{Reviews (2): Last reviewed commit: "docs: update PR_DESCRIPTION.md with erro..." | Re-trigger Greptile}

[greptile-apps]

Model prompts still lose tool names

When a persisted session is resumed or receives another turn, _build_prompt_messages() reconstructs rows without the ID-to-name map, so tool responses sent through LiteLLM still omit their function names and the targeted Gemini matching error remains reachable.