Stop silently swallowing script load/update/on_event errors

[diego3]

Summary

WorldScene.updateScripts (frameengine/view/scene/world_scene.go) discarded every error from
DoFile/DoString and CallScriptUpdate with a bare continue — no log, nothing. The three
CallOnEvent call sites (script-emitted events, BeginContact, EndContact) threw their error
away with _ =. Practical effect: a Python/Lua script with a syntax error or a runtime exception
left its entity silently frozen, with zero signal anywhere about why. Worse, a failed load never
marked loadedScripts[s.Path] = true, so the engine retried loading the same broken file every
frame, forever, just as silently.

Changes

• A load failure (DoFile/DoString/fs.ReadFile) is now logged once and the script is marked
  permanently failed (scriptLoadFailed), which also stops the per-frame retry loop — one fix
  covers both the missing log and the wasted repeated work.
• A CallScriptUpdate failure is logged only on the success→failure transition (and again on
  recovery), not on every frame it keeps failing — keeps this compliant with CLAUDE.md's "no
  per-frame logging" rule even for a script that errors continuously.
• The three CallOnEvent call sites (ScriptEmitted, BeginContact, EndContact) now log their
  error instead of discarding it. These fire per-event, not per-frame, so no rate-limiting is
  needed there.
• New test file world_scene_script_errors_test.go: a minimal script.Engine test double
  (per CLAUDE.md's "mock interfaces, not concrete types" convention) exercising updateScripts
  directly, covering: a failing load isn't retried every frame, a load failure is logged, an
  update failure logs once (not once per frame) while it keeps failing, and recovery is logged.

A real bug this immediately surfaced

Running games/demo1 live during validation now prints:

script: update(0.0167) failed in "games/demo1/scripts/knight_gamelogic.lua": script: global "update" is not a function (got nil)

games/demo1/scripts/knight_gamelogic.lua is an empty (0-line) stub. LuaEngine.CallScriptUpdate
calls the named function unconditionally without checking it's defined first — unlike
PythonEngine, which no-ops when the function is missing (per the Engine interface's own
documented contract: "Returns nil if the function is not defined (not an error)"). This predates
this change — the error was always being returned and always being silently discarded — so it's
not a regression from this PR. Left as-is deliberately: demo1's Lua scripts are legacy per
CLAUDE.md ("do not update or maintain them"), and fixing the Lua/Python behavioral inconsistency
itself is a separate, unrelated change. Flagging it here since it's now visible for the first
time. games/metalslug_demo (Python, the actively-developed game) runs clean with no such errors.

Validation

• go build ./..., go vet ./..., gofmt -l . — clean
• go test ./... — all pass; go test -race ./... — same result, except the pre-existing,
  unrelated box2d-go checkptr failure under -race (reconfirmed not caused by this change)
• New tests pass in isolation and as part of the full suite
• GOOS=js GOARCH=wasm go build — clean
• Live smoke run of both games/demo1 and games/metalslug_demo — see finding above; metalslug
  clean, demo1 now surfaces its pre-existing dead script (expected, not a regression)

---

Generated by Claude Code

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.