fix(firmware): raise the httpd stack for OTA image validation

[clonea1]

Adopted from upstream PR #1594 (Juan Kuscevic, open since 2026-08-11).

The OTA upload handler runs esp_ota_end() -> esp_image_verify() on the httpd
task's own stack, and HTTPD_DEFAULT_CONFIG allocates 4 KB. That overflows at
the very end of an upload: the transfer completes, validation panics, and the
node reboots into the old image. The symptom reads as 'the update did not
take' rather than as a crash, which makes it hard to attribute.

12 KB costs one page of RAM.

---

Stacked on #1827 (remote-config) — this branch carries that commit too, so review only the last two here, or merge #1827 first and this shrinks.

Rebased onto 33a9e908 (2026-09-14)

Two conflicts, both against files #1902 added at the same insertion points.
Both resolved additively:

• main/main.c — #include "serial_onboarding.h" and #include "config_api.h"
  both kept.
• main/CMakeLists.txt — serial_onboarding.c / serial_onboarding_protocol.c
  kept, config_api.c added. The branch's old side of this hunk re-listed
  main.c, csi_collector.c, stream_sender.c and nvs_config.c a second
  time in SRCS — a defect from an earlier resolution, not intent. SRCS now
  has 27 entries and no duplicates.

ota_update.c diffs as text again. The middle commit carried two literal
NUL bytes where '\0' was meant (s_rollback_reason[0] = '\0';, twice — a
lost backslash), which made git classify the file as binary: that commit and
this PR's headline commit both rendered as "Binary files differ", so the
change this PR is named after could not be read on GitHub. The escapes are
restored at their origin. The tree at the tip is bit-identical to a plain
rebase (0c8ed84b) — this changes how the history reads, not what ships.

The ESP-IDF 6 lane needed #1827's cJSON fix. Rebasing exposed this: on
33a9e908 the Build XIAO C6 external antenna (ESP-IDF 6.0.2) job passes,
and with only the three original commits it failed with
Failed to resolve component 'json' required by component 'main': unknown name
— config_api.c needs cJSON, which is bundled through 5.x and moved to the
component manager in v6. #1827's fix(firmware): source cJSON from the component manager on ESP-IDF 6 is cherry-picked here so this branch is green
on its own; it is the same patch, so it collapses if #1827 merges first.

Verification

• make -C firmware/esp32-csi-node/test host_tests — ALL PASS, 0 failures.
  The branch touches no file under test/, so that run is a regression control,
  not new coverage; config_api.c and ota_update.c are compiled by the
  firmware CI jobs, not by the host suite.
• No commit in this PR touches anything outside firmware/esp32-csi-node/.

Known-red check, and it is not clean-inherited

Build firmware (esp32s3 / 8mb) fails its Verify binary size budget step.
The compile itself is green; the budget is a single global number that main
already exceeds.

	binary	1152 KB gate	over by
main @ 33a9e908	1,191,568 B	1,179,648 B	11,920 B
this branch	1,204,208 B	1,179,648 B	24,560 B

So the failure is inherited, but the byte counts do not match: this branch
adds 12,640 B of firmware (config_api.c plus the rollback path), so it would
trip the current global gate even on a green main. It is not over its actual
partition — ESP-IDF reports the smallest app partition as 0x200000 (2 MB) with
43% free. #1891 ("size the firmware budget to each variant's own app
partition") is what makes this measurable rather than global.

🤖 Generated with claude-flow