merge: architecture into loops/haskell — Phases 7-16 complete + Phases 17-19 planned

Brings the architecture branch (559 commits ahead — R7RS step 4-6, JIT
expansion, host_error wrapping, bytecode compiler, etc.) into the
loops/haskell line of work. Conflict in lib/haskell/conformance.sh:
architecture replaced the inline driver with a thin wrapper delegating
to lib/guest/conformance.sh + a config file. Resolved by taking the
wrapper and extending lib/haskell/conformance.conf with all programs
added under loops/haskell (caesar, runlength-str, showadt, showio,
partial, statistics, newton, wordfreq, mapgraph, uniquewords, setops,
shapes, person, config, counter, accumulate, safediv, trycatch) plus
adding map.sx and set.sx to PRELOADS.

plans/haskell-completeness.md gains three new follow-up phases:
- Phase 17 — parser polish (`(x :: Int)` annotations, mid-file imports)
- Phase 18 — one ambitious conformance program (lambda-calc / Dijkstra /
  JSON parser candidate list)
- Phase 19 — conformance speed (batch all suites in one sx_server
  process to compress the 25-min run to single-digit minutes)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

plans/haskell-completeness.md

@@ -311,6 +311,54 @@ No OCaml changes are needed. The view type is fully representable as an SX dict.
     handler returns 0.
   - `trycatch.hs` — `try` pattern: run an action, branch on Left/Right.
 
+### Phase 17 — Parser polish
+
+Real Haskell programs use these on every page; closing the gaps unblocks
+larger conformance programs and removes one-line workarounds in test sources.
+
+- [ ] Type annotations in expressions: `(x :: Int)`, `f (1 :: Int)`,
+  `return (42 :: Int)`. Parser currently rejects `::` in `aexp` position;
+  desugar should drop the annotation (we have no inference at this layer
+  yet, so it's a parse-only pass-through).
+- [ ] `import` declarations anywhere at the start of a module — currently
+  only the very-top-of-file form is recognised. Real test programs that
+  mix prelude code with `import qualified Data.IORef` need this.
+- [ ] Multi-line top-level `where` blocks (`where { ... }` with explicit
+  braces and semicolons, in addition to the layout-driven form).
+- [ ] Tests for the above in `lib/haskell/tests/parse-extras.sx` (≥ 8).
+
+### Phase 18 — One ambitious conformance program
+
+Pick something nontrivial that exercises feature interactions the small
+suites miss; this is the only way to find unknown-unknown bugs.
+
+- [ ] Choose a target. Candidates:
+  - **Tiny lambda-calculus interpreter** (~80 LOC): parser, eval, env,
+    test cases. Stresses ADTs + records + recursion + `IORef` for state.
+  - **Dijkstra shortest-path** on a small graph using `Data.Map` +
+    `Data.Set`. Stresses Map/Set correctness end-to-end.
+  - **JSON parser** (subset): recursive-descent, exception-on-error,
+    `Either ParseError Value` results. Stresses strings + Either + try.
+- [ ] Adapt minimally; cite source as a comment.
+- [ ] Add to `conformance.conf`; verify scoreboard stays green.
+
+### Phase 19 — Conformance speed
+
+The full suite re-pays the ~30 s cold-load cost per program; 36 programs ⇒
+~25 minutes. Driving them all through one sx_server session would compress
+that to single-digit minutes.
+
+- [ ] In `conformance.sh` (and/or `lib/guest/conformance.sh`), batch all
+  suites into one process: load preloads once, then for each suite emit
+  an `(epoch N)` + `(load …)` + `(eval read-counters)` + `(eval reset-
+  counters)` block. Aggregate the per-suite results from the streamed
+  output.
+- [ ] Make sure a single failing/hanging suite doesn't poison the rest —
+  per-suite timeout via a server-side guard, or fall back to per-process
+  on timeout.
+- [ ] Verify the scoreboard output is byte-identical to the old per-process
+  driver, then keep the per-process path as `--isolated` for debugging.
+
 ## Progress log
 
 _Newest first._