lua: string.dump stub; diagnosed calls.lua fat-undef as at line 295

plans/lua-on-sx.md

@@ -82,6 +82,7 @@ Each item: implement → tests → tick box → update progress log.
 
 _Newest first. Agent appends on every commit._
 
+- 2026-04-25: lua: scoreboard iteration — `string.dump` stub (returns string-of-fn). Diagnosed calls.lua: file ENDS with `return deep` (line 295), which makes my `lua-has-top-return?` correctly return true, triggering the guard, which scopes user defines and breaks loadstring's lexical capture of `fat`. The fix would require either rewriting top-level returns into a different mechanism (post-processing the AST) or implementing a real `_G` global table for global assignment/lookup. Both larger.
 - 2026-04-25: lua: scoreboard iteration — `math.mod` (Lua 5.0 alias for `fmod`), `math.frexp` (mantissa/exponent split), `math.ldexp` (m·2^e). math.lua moves past line-103 `math.mod` call. Timeouts 6→3, asserts 5→7. 393/393 green.
 - 2026-04-25: lua: scoreboard iteration — `unpack(t, i, j)` now treats explicit nil for `i` or `j` as missing (defaults to 1 and `#t`). vararg.lua-style `unpack(args, 1, args.n)` works when `args.n` is nil. Asserts 4→5, timeouts 8→6 (more tests reach assertions instead of timing out). 393/393 green.
 - 2026-04-25: lua: scoreboard iteration — extended `string.format`. New `%x`/`%X`/`%o` (hex/octal), `%c` (codepoint→char via `lua-char-one`), `%q` (basic quote), width N (`%5d`), zero-pad (`%05d`), left-align (`%-5d`), `%.Ns` precision. Helpers `lua-fmt-pad` and `lua-fmt-int-base`. 393/393 green (+6 format tests).