Remote agent host: fix model picker on new session page

[roblourens]

The model picker on the new session page didn't show models for remote agent host sessions. Two issues:

1. Models never populated: The remote agent host server initially reports agents with 0 models (before authentication). After auth succeeds, the local agent host calls refreshModels() to trigger a root state update with the full model list — but the remote agent host contribution was missing this call. Added refreshModels() after _authenticateWithConnection and in _authenticateAllConnections, matching the local agent host pattern.

2. Model picker not visible: The localModelPicker action's when clause only matched the default Copilot provider (activeSessionProviderId == 'default-copilot'). Widened it to also match remote agent host providers via ContextKeyExpr.regex(ActiveSessionProviderIdContext.key, /^agenthost-/). Also made getAvailableModels dynamic — it now derives targetChatSessionType from the active provider's untitled session resource scheme instead of hardcoding AgentSessionProviders.Background.

Also: Restores two changes that were accidentally reverted by a bad merge in an earlier PR — the delegate.setModel fix (#305345) and multi-select context menu bridge (#306332).

(Written by Copilot)

[Copilot]

Pull request overview

This PR fixes a temporal-coupling issue in the chat language model registry where some providers could populate models before being registered, causing the service to miss the update and never surface those models (notably impacting remote agent host sessions and the model picker).

Changes:

• Always resolve language models when a provider is registered (instead of only when stored preferences exist).
• Add regression tests covering eager resolution without stored preferences and the “models updated before registration” scenario.
• Use ActiveSessionProviderIdContext.key in the remote agent host context expression and improve session context-menu bridging to support multi-selection.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
src/vs/workbench/contrib/chat/common/languageModels.ts	Makes provider registration always kick off model resolution to avoid missed pre-registration updates.
src/vs/workbench/contrib/chat/test/common/languageModels.test.ts	Adds/adjusts tests to validate eager model resolution behavior and cover the regression scenario.
src/vs/sessions/contrib/copilotChatSessions/browser/copilotChatSessionsActions.ts	Uses the centralized context key and improves command bridging to forward multi-selection session contexts.

Comments suppressed due to low confidence (1)

src/vs/workbench/contrib/chat/common/languageModels.ts:1006

• registerLanguageModelProvider now always kicks off _resolveAllLanguageModels asynchronously. If the returned IDisposable is disposed (provider unregistered) while that async resolution is still in-flight, the queued work can still complete and repopulate _modelCache after unregistration (because _resolveAllLanguageModels captures provider before the sequencer and doesn’t re-check that the provider is still the currently registered one before applying results). Consider adding a guard/version check inside the queued task (e.g., re-read this._providers.get(vendorId) and ensure it matches the provider instance that started the resolve) so stale resolves are discarded after unregistration.

		this._providers.set(vendor, provider);

		// Always resolve models on registration so callers don't need to
		// worry about firing onDidChange after registering the provider.
		this._resolveAllLanguageModels(vendor, true);

		const modelChangeListener = provider.onDidChange(() => {
			this._resolveAllLanguageModels(vendor, true);
		});