fix: Base AT-SPI ENABLED/SENSITIVE states on the disabled flag

[dylanpulver]

NodeWrapper::state computed STATE_ENABLED and STATE_SENSITIVE in the else arm of the read-only branch (adapters/atspi-common/src/node.rs:376-380), so neither ever consulted is_disabled. Measured on clean main by calling state() directly:

role=Button    disabled=true   => ENABLED=true  SENSITIVE=true  READ_ONLY=false
role=Button    disabled=false  => ENABLED=true  SENSITIVE=true  READ_ONLY=false
role=Link      disabled=true   => ENABLED=true  SENSITIVE=true  READ_ONLY=false
role=MenuItem  disabled=true   => ENABLED=true  SENSITIVE=true  READ_ONLY=false
role=TextInput read_only=true  => ENABLED=false SENSITIVE=false READ_ONLY=true

So a disabled node whose role doesn't support aria-readonly (Button, Link, MenuItem, Tab, ...) is indistinguishable from the enabled one, and a read-only-but-enabled control loses both states.

Core-AAM 1.2 §3.5.2.27 maps aria-disabled="true" to ATK/AT-SPI "State: STATE_ENABLED not exposed" and §3.5.2.28 maps aria-disabled="false" to STATE_ENABLED; the aria-readonly="true" row (§3.5.2.77) withdraws only STATE_EDITABLE. The other four adapters already derive the platform enabled state from is_disabled — macos/src/node.rs:660, windows/src/node.rs:535, android/src/node.rs:35, ios/src/node.rs:144 — and #474 was that same change for macOS.

The fix leaves the read-only branch exactly as it was and gates Enabled | Sensitive on !is_disabled(). Behaviour changes in the two cells above and nowhere else.

cargo +1.85 test -p accesskit_atspi_common (the MSRV toolchain CI uses): 7 passed on clean main, 11 passed here, 13 with --all-features. cargo fmt --all -- --check, cargo clippy -p accesskit_atspi_common --all-features --all-targets -- -D warnings and cargo doc with RUSTDOCFLAGS=-D warnings all clean locally.

The new tests discriminate: reverting the source alone fails 2 of the 4; the narrower } else if !state.is_disabled() { still fails the read-only one; dropping the read-only branch instead fails both READ_ONLY assertions. enabled_node_is_enabled_and_sensitive passes in every one of those trees.

Not tested: I have no Linux machine or Orca here, so this is verified at the accesskit_atspi_common translation layer only, not end to end through accesskit_unix against a live AT.

This change was prepared with AI assistance (Claude Opus 5).

[DataTriny]

Closing in favor of #788.

Please @dylanpulver next time consider opening an issue instead of submitting such contribution. Thanks.

[dylanpulver]

Understood, and noted for next time — I'll open an issue first here rather than arriving with a patch.

Glad the fix landed in #788.