feat: experimental `@st` structure definition v0

[idoros]

This PR is the first step in our plan for an improved way to define a stylesheet structure.

Changes in this PR add the ability to define nested pseudo-elements in a single stylesheet and focus on new syntax for defining classes, pseudo-elements, and pseudo-states. The new syntax allows to define deep nested pseudo-elements in the same stylesheet and currently enforce limits that exist in stylable (will be loosed in the future):

• classes with extend + global mapping that doesn't effect the type
  • @st .x :is(.y) => :global(.y)
• pseudo-element with mapping (no extend)
  • @st ::x => .y

More information on this effort can be found in the proposal.

Changes

Refactor parts handling

Up until now pseudo-elements were registered directly by css-class/ st-custom-selector features. This PR add a higher level feature st-structure to handle the previous auto parts mode (now called legacy-flat mode) and the new structure mode.

• st-structure checks the mode on analyze start and register parts according to mode
  • legacy-flat: set parts on .root class (from analyzed pseudo-elements and classes)
  • structure: analyze @st ::part definitions and register to the parent class or pseudo-element (nested pseudo-element is a new feature)

New @st definition

syntax: @st .class

• register top level @st .class to class-symbol[-st]
  • [x] error on non class param (remove in v3)
  • [x] error on definition not at top level
  • [x] collect => selector mapping (allow only global mapping for now)
  • [x] error on imported definition override
• inheritance (extends) - @st .class :is(<class-selector>)
  • [x] register top level class extended reference to class-symbol[-st-extends] - allow a single class extension
  • [x] transform inherited parts and states
• opt-in to new api
  • [x] console.warn about experimental feature
  • [x] when @st is used st-part ignores legacy parts (registered from @custom-selector and standalone css-class)
  • [x] no stylesheet implicit root class when using the new @st statement
  • [x] error on -st-extends/-st-states/-st-global (on class defined with @st .c<class>

Nested pseudo-state

syntax: @st :state;

• [x] register to class-symbol[-st-states]
  • transform selector
• [x] diagnostics
  • state limitations (no dash prefix, reserved natives)
  • out of context state definition
  • error on state re-declare (NEW JUST FOR @st :state)
• language service
  • [x] suggest nested state completion - .class:state::part \ .class::part:state

Nested pseudo-element

syntax: @st ::part => mapped-selector

• [x] top-level class parts
• [x] deep parts - new PartSymbol
• [x] pseudo-state definition
• [x] accept nest-selector (&) in mapping (similar to .root.x in @custom-selector)
• diagnostics
  • [x] error on redefine
    • including hard-coded native APIs (solved with @st-override in v2)
  • [x] error missing selector mapping
  • [x] error on multiple mapped selectors - for now this can be achieved by wrapping the selectors with :is() or &is()
  • [x] error out-of-context definition (directly in top level @st .class or @st ::pseudo-element)
• language service
  • [x] suggest nested parts completion - .class::partA::partB

Technical debts

• decide if to set @st .x class as root? maybe provide a way to opt-in/out?
• check why the default formatter makes mess of @st ::part => [mapped-selector]
• change css-class->st-symbol redeclare api to handle top level class in structure mode (error on declare)
• lsp for new @st definition
  • top level @st completion
  • class definition completion
    • class name: @st . / @st .<known-undefined-class>
    • extend class: @st .class :is(.base)
    • global map: @st .class => :global(<selector>)
  • pseudo-element completion - @st ::part => <selector>;
  • pseudo-state completion - @st :<state-def>;
  • go to definition for new definitions