Render HiGlass viewer inside Shadow DOM (#208)

* Render HiGlass viewer inside Shadow DOM

HiGlass v2.2.2 exports its CSS as `hglib.CSS`, which lets us encapsulate
viewer styles using the Constructable Stylesheets API and a Shadow DOM.
This prevents HiGlass styles from leaking into (or being affected by)
the host notebook environment.

The wheel event handler now also calls `preventDefault()` in addition to
`stopPropagation()`. With the shadow DOM boundary, `stopPropagation()`
alone wasn't sufficient to prevent page scrolling in marimo since
composed events cross the shadow boundary.

* Upgrade to higlass v2.2.3 for type export fix

* Fix type checking and linting

* ci: Upload smoke test failure screenshot as artifact

* Fix smoke test to check shadow root children

---------

Co-authored-by: Nezar Abdennur <nabdennur@gmail.com>

Merge pull request #201 from higlass/push-kwwowrxnvpmx

Inline esm.sh URLs in widget.js

Merge pull request #193 from higlass/kerpedji/sequence-logo-and-multi…

…vec-docs

fix,docs: Updated docs for multivec tracks and added a literal for sequence-log…

Add doc for chromsizes grid overlay (#187)

fix: Resolve Jupyter server tile sources in combined tracks (#186)

feat(widget): Batch tile requests within an animation frame (#181)

* feat(widget): Batch tile requests within an animation frame

HiGlass's built-in data fetching logic is designed to optimize API calls
by consolidating requests based on time, ID, and server. Our custom
Jupyter data fetcher doesn't include these optimizations.

This PR introduces a `consolidator` helper, which performs a much simpler
form of batching. Since we assume a single server, this function only
batches tile requests that occur within the same animation frame and
submits them together. This helps reduce the number of comms calls and
deduplicates requests efficiently.

Initial testing suggests it feels quite responsive!

* Add `MAX_TILES_PER_REQUEST` const

* Use `Promise.withResolvers`

* Remove chunkIterable for now

feat: Add stub `higlass.server` with deprecation warnings (#178)