fix: CSS back to <style> for Safari/cascade reasons

[microbit-matt-hillsdon]

Reverts the storage mechanism introduced in #9611 (constructable stylesheets via adoptedStyleSheets) while keeping the per-root injection-site tracking that #9611 added for shadow-DOM support.

Motivations:

• Safari 15.4 compatibility. new CSSStyleSheet() and adoptedStyleSheets require Safari 16.4+
• Cascade order. adoptedStyleSheets apply after <style>/<link> elements in the document, so Blockly's defaults silently overrode host stylesheets. Prepending a <style> to the head (or to the shadow root) restores the pre-feat!: Allow using Blockly in web components/shadow DOM #9611 behavior where any author stylesheet declared later wins on specificity ties.

Trade-offs:

• Per-shadow-root CSS text is duplicated rather than shared via a single adopted sheet object. Negligible for typical use.
• Css.register() calls made after the first inject() no longer reach already-injected roots (same as feat!: Allow using Blockly in web components/shadow DOM #9611's behavior); subsequent inject() calls into other roots still pick them up. Web-component consumers can legitimately register late, so this is preferred to reinstating the pre-feat!: Allow using Blockly in web components/shadow DOM #9611 throw.

@gonfunko this is one possible take with the downsides noted above. Another option is to use both mechanisms (tag and adoptedStylesheets) either based on feature detection or based on the root. Then to fix the cascade issue consistently use "@layer blockly" for Blockly's CSS. That will require auditing of the !important use in Blockly as they would become impossible to override. If relying on feature detection note it is not sufficient to check for CSSStyleSheet being defined as that significantly predates constructable stylesheets.

Please feel free to make changes to this PR and/or replace it with one of your own.

The basics

• I validated my changes

I didn't format as it made a large unrelated diff whoops, I had outdated prettier, all good now.

The details

Resolves

Fixes #9876

Proposed Changes

Reason for Changes

Test Coverage

I checked the web component playground added in the previous PR.

Documentation

Additional Information

[gonfunko]

Thanks for putting this together, looks reasonable to me but I'll let Maribeth have a look before merging.

[gonfunko]

Thanks for putting this together, looks reasonable to me but I'll let Maribeth have a look before merging.