chore(repo): untrack Android Gradle cache, gitignore noise, bench TS-strip leveling (v0.5.293)

- Untrack 465 files under `android-build/.gradle/`, `android-build/build/`,
  and `android-build/app/build/` — Gradle daemon cache + AGP intermediates
  + APK outputs that churned on every build. Real source under android-build/
  (Kotlin, manifest, gradle scripts, gradle wrapper, resources, jniLibs)
  stays tracked.
- .gitignore: add the cache dirs above, plus `docs/examples/_reports/`
  (generated by scripts/run_doc_tests.sh, uploaded as a CI artifact),
  `/assets/` + `benchmarks/suite/assets/` (external game-project assets
  the user keeps adjacent for perry-ui-* manual testing — never source),
  and stray repro binaries `enum_repro`/`no_pragma_test`.
- json_polyglot/run.sh: precompile Node TS to .mjs (esbuild → npx-esbuild
  → tsc fallback) as untimed setup so Node isn't charged for
  --experimental-strip-types' per-launch parse cost. Perry is AOT and
  Bun strips natively; neither pays this. Falls back to
  --experimental-strip-types with a banner if no stripper is available.
- polyglot/bench.rs: FP-contract caveat block on bench_loop_data_dependent
  documenting the FMA-contract (Apple Clang, Go) vs no-contract (Rust,
  Swift, Perry, Node, Bun, Java) clustering on Apple silicon.
- tests/test_array_index_loop.sh: runner companion to the existing .ts
  regression test (depends on adjacent ../../bloom/jump checkout, skips
  cleanly if absent).

Author: proggeramlug

.gitignore

@@ -3,6 +3,28 @@ target/
 *.o
 *.app/
 
+# Android Gradle: caches and build outputs are regenerable. Source under
+# android-build/ that we DO track: build.gradle.kts files, gradle wrapper,
+# AndroidManifest.xml, Kotlin sources under app/src/main/java, and resources
+# under app/src/main/res. Everything below is daemon cache or compiler
+# intermediates that churn on every build.
+android-build/.gradle/
+android-build/build/
+android-build/app/build/
+android-build/local.properties
+android-build/.idea/
+android-build/captures/
+
+# Doc-example test reports (generated by scripts/run_doc_tests.sh,
+# uploaded as a CI artifact in test.yml — never source)
+docs/examples/_reports/
+
+# External demo assets (game project assets the user keeps adjacent for
+# perry-ui-* manual testing — not perry source). The publish.rs `assets/`
+# references are about the *consumer's* project layout, not perry's repo.
+/assets/
+benchmarks/suite/assets/
+
 # V2.2 per-module object cache (machine-local: keyed by opts + source hash,
 # but the cached .o files bake in host CPU features via clang -mcpu=native /
 # -march=native, so sharing across machines with different CPUs can produce
@@ -84,6 +106,8 @@ http_server
 searchbird_*
 image_conv
 issue42_noalloc
+enum_repro
+no_pragma_test
 
 # Test output
 test-parity/output/

CLAUDE.md

@@ -109,7 +109,7 @@ opt-level = "s"       # Optimize for size in stdlib
 opt-level = 3
 
 [workspace.package]
-version = "0.5.292"
+version = "0.5.293"
 edition = "2021"
 license = "MIT"
 repository = "https://github.com/PerryTS/perry"