feat: relax structured output parsing

[maksymbuleshnyi]

Note

Medium Risk
Touches core agent output parsing: strict=False will accept previously-invalid JSON (e.g., literal newlines), which can change what gets executed as tool inputs and may mask malformed model outputs.

Overview
Agent parsing is made more tolerant of imperfect LLM JSON. Agent._handle_structured_output_mode and Agent._handle_function_calling_mode now parse with json.loads(..., strict=False) (and the structured-output fallback JSONDecoder(strict=False)) so tool action_input can contain literal newlines and the first JSON object can be extracted from concatenated responses.

Prompts and tests are updated accordingly. The structured-output and function-calling instructions add stricter guidance about field ordering and newline escaping, and new unit tests cover literal-newline action_input plus the multi-JSON fallback path.

^{Reviewed by Cursor Bugbot for commit 575dee4. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit b5898ad. Configure here.}

[github-actions]

Coverage Report •

File	Stmts	Miss	Cover	Missing
dynamiq/nodes/agents
agent.py	693	135	80%	213–216, 221, 229, 234, 252–253, 271–272, 325–327, 488–497, 500–502, 517, 526, 535–536, 552, 555–556, 566–569, 574, 587, 600–601, 616, 628, 633, 649–650, 660, 663, 673, 713–715, 721–723, 837, 842, 892, 926, 928, 943–944, 951, 954, 979, 981–982, 998, 1023–1027, 1094, 1100, 1104, 1107, 1140, 1165, 1168, 1173, 1189, 1193, 1235, 1271–1272, 1298, 1312, 1321, 1324–1325, 1335, 1342, 1404–1405, 1409, 1411–1413, 1415, 1417, 1450, 1458–1459, 1461, 1464, 1467, 1518–1521, 1537, 1546, 1585, 1591–1597, 1603, 1605–1614, 1616–1617
TOTAL	26780	8960	66%

Tests	Skipped	Failures	Errors	Time
1548	1 💤	0 ❌	0 🔥	1m 54s ⏱️