Add DAOP explainer and estimateQoS() illustration with background blur demo

[jonathanding]

This PR introduces the Dynamic AI Offloading Protocol (DAOP) proposal for the WebNN API, along with a working JavaScript illustration.

DAOP proposes an estimateQoS() API that enables web applications to evaluate whether a specific AI model can run acceptably on the local device — without exposing raw hardware details or compromising user privacy. The API returns a coarse performance tier (e.g., "excellent", "fair", "poor"), allowing the application to decide between local execution and cloud offloading.

What's included:

• README.md — Full W3C-style explainer covering motivation, API design, privacy considerations, and alternatives
• daop-illustration/ — Browser-based polyfill demonstrating a potential approach to implement it, for illustration purpose
• Interactive background blur demo applying the DAOP workflow end-to-end: build weightless graph → estimate QoS → offloading decision → local execution

See the DAOP proposal issue (webmachinelearning/proposals#15) for background discussion.

[anssiko]

@jonathanding presented (thanks!) the DAOP explainer and estimateQoS() prototype at WebML WG Teleconference – 12 February 2026 and the group had a discussion.

As the next step, the group will review the explainer in this PR (preview). After adequate review time has passed (~2 weeks) and the feedback has been addressed, the group will merge this PR to establish a baseline for further refinement.