gpui-base: Add TextSelection for window-level text selection (#2730)

## Summary

- add renderer-neutral, window-level text selection to `gpui-base`
- migrate `TextView` to the shared selection engine while preserving
rich-text parsing, source copy, virtualization, focus, and scrolling
behavior
- isolate background, Dialog, and Sheet selection with opaque per-window
scopes
- keep the existing `gpui-component` selection methods as deprecated
forwarding shims
- add a selectable-text showcase, integration documentation, and
lifecycle/compatibility coverage
- correct the release workflow to publish the renamed `gpui-base` crate

## API design

The public API separates window coordination from renderer-owned state:

- `TextSelectionLayer` is mounted once per window and retains the window
selection state and pointer handlers.
- `TextSelection` provides window-level query and control operations:
`selected_text`, `has_selection`, `clear`, `end`, and `activate_scope`.
- `TextSelectionHandle` is retained by each selectable participant and
owns its snapshot, callbacks, local-selection state, and plain-text
projection.
- `TextSelectionRegistration` reports per-frame geometry, document
order, scroll offset, and scope.
- `TextSelectionRun` and `TextSelectionProjection` provide UTF-8-safe
range projection for ordinary laid-out text.
- `TextSelectionSnapshot`, `TextSelectionEndpoint`,
`TextSelectionCoverage`, and `TextSelectionContentKey` support custom
and virtualized renderers without exposing internal window state.
- `ElementExt::text_selection_scope` marks modal or otherwise isolated
subtrees.

The engine keeps participant identity stable, uses explicit logical
ordering instead of paint or map order, stores endpoints in
participant-relative content coordinates, expires registrations by
frame, and keeps window registries weak so closed windows release their
selection state.

## Compatibility

Existing `TextView` behavior remains covered for click/drag and Shift
extension, multi-click word/line selection, cross-participant copy,
plain/source formats, virtualized blocks, focus, auto-scroll,
drag-and-drop suppression, layout changes, modal scopes, and
multi-window cleanup.

Applications using `gpui_component::Root` receive the
`TextSelectionLayer` automatically. The former `Root` and `WindowExt`
selection methods remain available as deprecated synchronous forwards to
the single base-owned state.

## Verification

- `cargo test -p gpui-base text_selection` — 43 passed
- `cargo test --workspace --all-targets` — passed
- `cargo check --workspace --all-targets` — passed
- `cargo clippy --workspace --all-targets -- --deny warnings` — passed
- `cargo fmt --all -- --check` — passed
- `git diff --check` — passed
- `bun run build` in `website` — passed

## Status

Kept as draft for maintainer API review.

---------

Co-authored-by: Codex <codex@openai.com>