feat(coap): envelope ↔ CoAP payload translation

[rodrigopex]

Summary

Phase 2 of the CoAP-frontend umbrella (#27). Adds the pure, transport-free envelope ↔ CoAP payload translation layer plus the return_code → CoAP class mapping table. No dependency on coap_service, sockets, or zbus — the helpers operate on raw buffers and pre-initialised struct coap_packet instances, so they are unit-testable in isolation.

Wire format

Request and response bodies are nanopb-encoded protobuf messages described by the same pb_msgdesc_t descriptors the per-zephlet _interface.c already references. Content-Format ZEPHLET_COAP_CT_NANOPB (65001) tags outbound payloads; the verbatim raw POSIX errno rides every response via custom option ZEPHLET_COAP_OPT_ERRNO (65052) so the (lossy) class mapping never erases the precise handler return code.

What landed

• frontends/coap/include/zephlet_coap_translate.h — public API: zephlet_coap_map_return_code, _decode_request, _encode_response.
• frontends/coap/zephlet_coap_translate.c — implementation.
• frontends/coap/include/zephlet_coap_types.h — struct zephlet_coap_method gains req_max_size / resp_max_size (size_t) — codegen-emitted, see below.
• codegen/templates/zephlet_coap_interface.c.jinja — emits the two new fields from nanopb's compile-time <MSG>_SIZE macros (zero for Empty).
• frontends/coap/CMakeLists.txt — adds the new TU to the frontend library.
• tests/coap_translate/ — new twister target (harness: ztest, mps2/an385 + native_sim) with a tiny local test_msg.proto. Compiles zephlet_coap_translate.c directly so the test does not have to enable CONFIG_ZEPHLETS_COAP (no networking/sockets pulled in).

Why codegen-derived per-method bounds vs. a Kconfig cap

The decoder enforces len > m->req_max_size → -EMSGSIZE (→ 4.13). The bound is the exact nanopb-computed maximum encoded size of the method's request message, so an Empty-typed method rejects any inbound body through the same path (req_max_size == 0). No single tuning knob to keep in sync with the protos — adding a method whose nanopb size exceeds the old default would have been a silent regression.

return_code → CoAP class table

rc	class
0 (with payload)	2.05 Content
0 (no payload)	2.04 Changed
-EINVAL	4.00 Bad Request
-ENODEV	4.04 Not Found
-ENOSYS	4.05 Method Not Allowed
-EALREADY	4.09 Conflict
-EMSGSIZE	4.13 Request Entity Too Large
-EBUSY / -EAGAIN	5.03 Service Unavailable
-ETIMEDOUT	5.04 Gateway Timeout
-ENOMEM / any other negative	5.00 Internal Server Error

-EMSGSIZE → 4.13 was the only addition vs. the plan's locked-decisions table — confirmed EMSGSIZE is defined in zephyr/lib/libc/minimal/include/errno.h:106.

Test results (local, mps2/an385 + QEMU 10.0.2)

• New tests/coap_translate/: 22/22 ztest cases pass in 0.014s.
• Existing tests/codegen/ pytest suite: 6/6 pass (codegen-template change validated).
• Existing app integration suite (tick / ui / tampering): 23/23 pass.
• App build with CONFIG_ZEPHLETS_COAP=n (default): clean — =n section-hash gate inherits unchanged because the new TU only compiles under =y.

Out of scope (deferred to later phases)

• URI parsing / dispatch (Phase 3 — feat(coap): Phase 3 — RPC dispatch (network, unary) #31).
• Discovery (/.well-known/core) — Phase 4 (feat(coap): Phase 4 — discovery (/.well-known/core) #32).
• Event bridge / Observe — Phase 5 (feat(coap): Phase 5 — event bridge (CoAP Observe) #33).
• Hardening + DTLS — Phase 6 (feat(coap): Phase 6 — hardening + optional DTLS #34).
• macOS QEMU+vmnet for the Phase 0 functional pytest harness — orthogonal; file a follow-up if desired.

Test plan

• west twister tests/coap_translate green (mps2/an385).
• tests/codegen/ pytest green (codegen template regression).
• App build clean with CONFIG_ZEPHLETS_COAP=n.
• App just test green (no integration-test regression).

Closes #30