Refactor model selection logic into separate file. Add tons of tests

[lramos15]

The number of user reports about the model picker incorrectly switching from under them is quite high.

This extracts the logic into its own file since it's a bit complex and also adds a ton of tests

Some bugs include

• [Bug] Chat Model Picker Broken For Custom Model Providers #299172
• Model dropdown always reverts back to my BYOK model #296236
• Switches to byok model #286285
• Agent model resets to GPT-4.1 after restart instead of persisting Claude selection #291940

[Copilot]

Pull request overview

This PR refactors chat model selection/reset logic out of ChatInputPart into a dedicated chatModelSelectionLogic.ts module and adds extensive unit test coverage to reduce “model picker switches out from under me” issues reported across multi-window/session scenarios.

Changes:

• Added chatModelSelectionLogic.ts with reusable helpers for model filtering, default selection, persistence restore, sync-state reconciliation, and cache merging.
• Updated ChatInputPart to use the extracted helpers (model list changes, persisted restore, sync-from-state, validation/reset, cache merge, default selection).
• Added a large new test suite covering mode/session/inline constraints, persistence semantics, cache merge behavior, and race-like sequences.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
src/vs/workbench/contrib/chat/browser/widget/input/chatModelSelectionLogic.ts	New module encapsulating model selection, filtering, reset, and cache-merge decision logic.
src/vs/workbench/contrib/chat/browser/widget/input/chatInputPart.ts	Switched in-widget logic to the new helpers; refactored persistence restore and sync/reset paths.
src/vs/workbench/contrib/chat/test/browser/widget/input/chatModelSelectionLogic.test.ts	New, comprehensive unit test suite for the extracted logic and key scenarios/races.

Comments suppressed due to low confidence (2)

src/vs/workbench/contrib/chat/browser/widget/input/chatModelSelectionLogic.ts:32

• The session-type check hardcodes the string 'local'. Since the codebase already defines localChatSessionType (in chatSessionsService.ts), it would be more robust/maintainable to depend on that constant here to avoid future drift.

	if (sessionType && sessionType !== 'local') {
		return models.filter(entry =>

src/vs/workbench/contrib/chat/browser/widget/input/chatInputPart.ts:953

• There are now several repeated blocks that rebuild allModels via getLanguageModelIds().map(lookupLanguageModel(...)) (e.g. _syncFromModel, checkModelSupported, hasModelsTargetingSessionType, isModelValidForCurrentSession). Besides allocations, this duplication makes it easy for these call sites to diverge (e.g. around cache inclusion). Consider extracting a single helper (and reusing the same merged live+cache list where needed).

				const allModels = this.languageModelsService.getLanguageModelIds()
					.map(modelId => ({ identifier: modelId, metadata: this.languageModelsService.lookupLanguageModel(modelId)! }));
				const sessionType = this.getCurrentSessionType();
				const syncResult = resolveModelFromSyncState(state.selectedModel, this._currentLanguageModel.get(), allModels, sessionType, {
					location: this.location,
					currentModeKind: this.currentModeKind,
					isInlineChatV2Enabled: !!this.configurationService.getValue(InlineChatConfigKeys.EnableV2),
					sessionType,
				});
				if (syncResult.action === 'apply') {