feat(wasm): cwist_db in WASM build, EM_JS format-corruption fix, wasm docs (issue #93)

[gg582]

Part of #93 (Phase 1 + start of Phase 2). Draft.

What's here

• Makefile: src/core/db/db.c + lib/sqlite3/sqlite3.c added to WASM_SRCS - the ROADMAP claim that cwist_db_open_memory()/cwist_db_serialize() work under WASM is now actually true (gap 1). Compiles clean with emcc.
• tests/wasm_smoke.c: cwist_db round trip - create + insert, serialize to an image blob, reopen read-only via cwist_db_open_memory(), query the row back. Also fixes JS that the tree-wide clang-format pass corrupted (=> rewritten to = >): EM_JS stringifies its raw argument, so brace-block bodies are now wrapped in clang-format off/on or make format silently breaks the WASM build.
• include/cwist/wasm/typedarray.h: same clang-format protection for CWIST_WASM_INSTALL_VIEWS.
• docs/api/wasm.md: initial documentation (gap 4) - build, scope, dispatch_memory pattern, TypedArray helpers, db round trip, session caveats.
• .github/workflows/wasm.yml: Emscripten 5.0.0 build + wasm-smoke on every push/PR to main/dev (gap 2). Includes a workaround for the actions/cache emsdk exec-bit loss; NODE is now overridable in the Makefile.

Verified

• make wasm + make wasm-smoke locally (Emscripten 5.0.0 + node), make format idempotent wrt EM_JS bodies.
• WASM CI job green on this PR.

Still open (later PRs)

• npm/release packaging, first-party JS wrapper (Phase 2)
• WASI target, streaming, session persistence model, example app (Phases 3-4)

[gg582]

@Bhumika-1432006
Thanks for picking up issue #93 from Discord. Reviewed the WASM changes themselves (Makefile, wasm.yml, wasm_smoke.c, typedarray.h, wasm.md) and they look solid; the Emscripten CI job is a real gap-closer.

One structural request before this leaves draft:

Scope the diff to the 5 WASM files. The branch currently carries a merge of #175 and a main-side benchmark report refresh, so the diff is 14 files and includes README/benchmark numbers from main. Please rebase onto current dev (or drop the merge commits) so the PR only touches the WASM work. Mergeable is CLEAN either way, but merging as-is would land main's benchmark numbers into dev's README.

When you do, one thing to keep: the FIN=1 orphan CONTINUATION check from #175 that rides along in this branch is genuinely missing on dev (an orphan CONTINUATION with FIN=1 falls through both branches in cwist_websocket_receive() and is delivered as a complete message). If the rebase drops it, please split it into its own small PR and we will merge that first.

And a few follow-ups we would appreciate, roughly in priority order. Items 1 and 2 are small enough to fold into this PR; 3 and 4 can be separate issues if you prefer:

1. WASM binary size report. Adding src/core/db/db.c + lib/sqlite3/sqlite3.c pulls all of SQLite into the wasm archive. A before/after size for libcwist_wasm.a and the linked wasm_smoke.wasm gives us the data to decide on a future opt-out or split build for users who do not need cwist_db.

2. Run make format-check in the wasm CI job. The EM_JS => corruption you fixed recurs silently the next time anyone runs make format. The new job builds and smokes, but does not guard the formatter; adding one line to wasm.yml closes that loop for good.

3. Turn the "session caveats" in wasm.md into measured behavior. The doc lists session persistence limits, but nothing tests what actually works vs fails under the WASM dispatch model. A short writeup (or a few smoke assertions) of observed behavior would anchor the Phase 3 session design.

4. Phase 2 handoff. npm packaging and a first-party JS wrapper are the bulk of the next phase. Let us know if you want to keep going; if not, even a short API sketch (what to expose, module shape) would let someone else pick it up.

Happy to re-review quickly once the branch is rebased.