Python: Refactor workflows to use SerializationMixin from DictConvertible

[moonbox3]

Motivation and Context

The workflow code uses a separate DictConvertible class with methods like to_dict(), from_dict(), to_json(), and from_json(). These methods are also available in the core library under SerializationMixin. So the code is aligned, workflows is moving to use SerializationMixin removing the need for DictConvertible.

This is a pure internal refactoring that maintains complete backwards compatibility while standardizing the codebase on SerializationMixin. Users won't need to change any code, and all existing serialized workflows/checkpoints will continue to work. Users will have access to additional optional features like exclude and exclude_none parameters in to_dict(), which is an enhancement, and a breaking change.

Description

• Closes Python: Implement SerializationMixin in place of DictConvertible #1797

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? If yes, add "[BREAKING]" prefix to the title of the PR.

[markwallace-microsoft]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework/_workflows
_edge.py	219	26	88%	37, 59–60, 62, 414–415, 420, 424–429, 431–433, 543, 657, 664–666, 713, 741, 769, 823, 837
_executor.py	138	6	95%	203, 319, 334, 336, 451, 461
_magentic.py	931	282	69%	70–79, 84, 88–99, 323, 328, 345, 347, 362, 370–379, 457, 461, 475, 483, 498, 584, 597, 614, 623–624, 626–628, 630, 641, 691, 708–712, 715–719, 799–802, 805–809, 811–813, 820, 859, 906, 942–944, 946, 1013, 1086, 1088–1089, 1097, 1123–1124, 1133, 1142–1144, 1163, 1174, 1184–1185, 1187–1188, 1207, 1260, 1280, 1283, 1311–1315, 1317, 1319, 1321, 1323–1327, 1333, 1339, 1341, 1345, 1348–1350, 1353, 1357–1358, 1361–1363, 1365, 1367, 1375–1378, 1380, 1384–1385, 1388–1391, 1393–1394, 1400–1402, 1405–1406, 1411–1412, 1420, 1428, 1443, 1455, 1467–1470, 1499–1500, 1505–1507, 1538, 1564, 1579, 1595, 1612, 1682, 1688–1689, 1703, 1705, 1708, 1710–1711, 1714–1715, 1719, 1722, 1743, 1780–1781, 1783, 1787–1789, 1805, 1816, 1826, 1870, 1875–1876, 2202–2203, 2207, 2222, 2227, 2230, 2284, 2295, 2306–2308, 2321–2322, 2327, 2338–2340, 2351–2353, 2365–2372, 2374–2375, 2383, 2391–2392, 2394–2396, 2398–2401, 2405–2413, 2417–2418, 2421–2425, 2427–2428, 2430–2433, 2435–2438, 2440–2442, 2444–2445, 2447–2449, 2464–2467, 2478–2481, 2493–2496, 2500
_model_utils.py	10	5	50%	12–16
_workflow.py	246	18	92%	94, 263–265, 267–268, 282, 306, 308, 401, 650, 684, 689, 692, 711–713, 778
TOTAL	12805	1905	85%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
1698	107 💤	0 ❌	0 🔥	33.489s ⏱️

[eavanvalkenburg]

so made one comment, that applies to many classes, the design calls for some additional leeway in the init, that parses dicts and calls from_dict on those items, so that you do not need to subclass from_dict for each (it does the inverse OOTB in to_dict). That also ensures the dependency injection can work across the board, which will probably be usefull for declarative support down the line.

[eavanvalkenburg]

so in order for the serializaiton to work OOTB into nested classes, this would need to accept Sequence[Mapping[str, Any]] and then if the type is not an Edge, it should call Edge.from_dict(), that way you do not have to customize the from_dict for each class, to handle that behavior.