Add 2025.3.1 backwards-compat to ML models

[mcm001]

Description

In #1894 we decided for some reason not to be backwards compatible. But fixing this requires SSHing in and deleting files which is ugly. It's also not that much code to maintain compat. It also causes configs to be partially lost if Object Detection pipelines were saved (discovered via Discord report). In light of that discovery, changes were made to increase robustness to exceptions during config loading, including a more fine-grained try-catch.

Changes

• NeuralNetworkModelManager::loadModel will load legacy models from disk, and save to ConfigManager.getInstance().getConfig().neuralNetworkPropertyManager().addModelProperties(properties);; this saves them to disk
• Renames NeuralNetworkPropertyManager to NeuralNetworkModelsSettings
• Add ModelProperties string-only constructor to handle legacy serialization/deserialization from old ObjectDetection pipelines that just saved the name
• Move try-catch in SqlConfigProvider::loadCameraConfigs to be per-camera; one bad camera won't cause the rest to be skipped

Meta

Merge checklist:

• Pull Request title is short, imperative summary of proposed changes
• The description documents the what and why
• If this PR changes behavior or adds a feature, user documentation is updated
• If this PR touches photon-serde, all messages have been regenerated and hashes have not changed unexpectedly
• If this PR touches configuration, this is backwards compatible with settings back to v2025.3.2
• If this PR touches pipeline settings or anything related to data exchange, the frontend typing is updated
• If this PR addresses a bug, a regression test for it is added