examples: lisp — chunked bump arena for value/env cells

Replaces rt_alloc-per-cell (one 4 KiB mmap each) with arena_alloc
over 64 KiB chunks. test_huge peak under massif --pages-as-heap=yes
drops from ~525 MB to ~253 MB. Same lifetime semantics; remaining
bulk is per-call append() in eval's arg slice (rt_ensure still
mmaps page-per-call).

examples/lisp/CLAUDE.md

@@ -115,14 +115,17 @@ env where `od?` is unbound; the tie-back only adds the self-binding.
 
 ## Interpreter limitations (design, not bug)
 
-- **No GC.** Every cons / value / env frame is `mmap`'d via
-  `rt_alloc` and never reclaimed. A long REPL session leaks until
-  the process exits. The test_huge demo peaks at ~595 MB under
-  `--pages-as-heap=yes`. Per-top-level-form arena reset would cut
-  this by ~100× — see the closing note in `repl()` for the hook
-  point. Note: `rt_alloc` is one mmap syscall per call returning a
-  whole 4KB page, so even tail-recursive loops are bottlenecked on
-  allocation, not on Lisp work — `(spin 100000 0)` runs in ~4s.
+- **No GC.** Cons / value / env cells come from a single bump-
+  pointer arena (`arena_alloc`, 64 KiB chunks chained via `os.alloc`).
+  The arena never reclaims — long REPL sessions still leak — but
+  cells pack tightly instead of one cell per 4 KiB mmap. test_huge
+  peaks at ~253 MB under `--pages-as-heap=yes`, down from ~525 MB
+  pre-arena. The remaining bulk is `append(xs, av)` in eval's arg
+  loop: every apply allocates a fresh `[]*value` through `rt_ensure`,
+  which still hits `rt_alloc` page-per-call. Per-top-level-form
+  arena reset (perm/trans split + Cheney promote on define/set!) is
+  the next step — see the closing note in `repl()` for the hook
+  point.
 - **No bigints.** `i64` wraps silently on overflow. `(fact 21)`
   rolls over.
 - **Float printing is fixed `%.6f`.** `1.0` prints as `1.000000`.