Add long-context usage warning indicator

[dagmfactory]

Adds a long-context usage warning indicator to the agent input footer, with the client deciding visibility from server-provided threshold and input data.

• What: Thread the server's total_input_tokens (latest primary-agent call input tokens of the latest persisted charged/BYOK request) through the proto stream, GraphQL restore, and local persistence, and expose each model's longContextThreshold on LlmContextWindow.
• The footer warning shows when provider == OpenAI && total_input_tokens > threshold for the active model.
• Why: Signal to users when a conversation has entered OpenAI's long-context pricing tier, without the server computing a boolean — the client recomputes against the active model's threshold (e.g. on model switch).
• How:
  • crates/persistence/src/model.rs, app/src/ai/agent/conversation.rs: persist/apply total_input_tokens (nonzero overwrites, zero keeps previous, matching server merge semantics).
  • crates/graphql/src/api/..., crates/warp_graphql_schema/api/schema.graphql: totalInputTokens on conversation usage; nullable longContextThreshold on LlmContextWindow; LLMContextWindow.long_context_threshold on the client model metadata.
  • app/src/ai/blocklist/usage/mod.rs: LongContextWarningState now holds tokens + active model provider/threshold and derives visibility; UI rendering (icon, tooltip, yellow warning theme) is unchanged.

Counterparts:

• proto: Replace long_context_used with total_input_tokens on ConversationUsageMetadata warp-proto-apis#323 (total_input_tokens, pinned at 7de66cc8)
• server: warpdotdev/warp-server#11843

Linked Issue

• The linked issue is labeled ready-to-spec or ready-to-implement.
• Where appropriate, screenshots or a short video of the implementation are included below (especially for user-visible or UI changes).

Testing

• app/src/ai/blocklist/usage/mod_tests.rs: threshold boundary (at vs +1), no-threshold model (Auto), non-OpenAI providers hidden despite threshold, model-switch recompute (lower/higher/non-OpenAI), zero tokens, icon behavior.
• crates/persistence/src/model_tests.rs: total_input_tokens JSON round-trip and legacy-row default.
• cargo nextest targeted run (56 tests) passes; ./script/format run; cargo clippy --all-targets -- -D warnings clean on touched packages (warp, persistence, warp_graphql).

Screenshots / Videos

https://www.loom.com/share/e2ea70c8db7a4ce6ada8c766c5a425da

different themes:
https://www.loom.com/share/9d0d634e82314f39ba8e1eab0c69363e

Agent Mode

• Warp Agent Mode - This PR was created via Warp's AI Agent Mode

Plan: https://staging.warp.dev/drive/notebook/yJt0g7NfiwnSYuQdNfQAO1
Conversation: https://staging.warp.dev/conversation/4321295c-fe0b-4a19-9382-44738d7db792

Co-Authored-By: Oz oz-agent@warp.dev

[oz-for-oss]

@dagmfactory

I'm starting a first review of this pull request.

You can view the conversation on Warp.

I completed the review and no human review was requested for this pull request.

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

[oz-for-oss]

Overview

This PR adds long-context pricing-warning state, wiring from conversation usage/model metadata, a themed context-window chip warning, and persistence/API fields for total input token counts.

Concerns

• The warning visibility path does not consult the new LongContextWarning feature flag, so the new UI can appear outside the intended dogfood/flag rollout.
• This is a user-facing UI change, but the PR description does not include screenshots or a screen recording demonstrating the warning end to end. Please attach visual evidence for this change before merge.

Verdict

Found: 0 critical, 2 important, 0 suggestions

Request changes

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

[oz-for-oss]

⚠️ [IMPORTANT] This bypasses the new LongContextWarning feature flag, so any user with threshold metadata can see the UI outside the intended rollout. Gate the visibility on FeatureFlag::LongContextWarning.is_enabled().

Suggested change

	let show_long_context_warning = self.long_context_warning_state.is_visible();
	let show_long_context_warning = FeatureFlag::LongContextWarning.is_enabled()
	&& self.long_context_warning_state.is_visible();