fix(firmware): restore native C6 onboarding delivery

[proffesor-for-testing]

Outcome

Restores the Mac USB/JTAG onboarding contract and makes firmware releases consumable as exact artifacts.

Changes

• make USB Serial/JTAG the primary ESP32-C6 console so RUVIEW_HELLO_V1 reaches stdin;
• remove the ineffective /dev/secondary reader (ESP-IDF secondary consoles are output-only);
• replace the tracked C6 onboarding bundle so it identifies itself as 0.8.12;
• preserve PR fix(ci): size the firmware budget to each variant's own app partition #1891's partition-relative size budgets, allowing the S3 8 MB build to use its real 2 MB OTA slot while retaining 20% headroom;
• publish exact-run S3 8 MB, S3 4 MB, and C6 4 MB archives on dedicated vX.Y.Z-esp32 releases;
• document that automated vNNNN server releases such as v2655 are not firmware releases;
• add a static delivery contract to the existing host-test gate.

Verification

• red-first contract reproduced both faults: C6 was not configured for primary native USB and the tracked app embedded 0.8.11;
• make host_tests passes;
• local ESP-IDF 6.0.2 C6 build passes and generated sdkconfig contains CONFIG_ESP_CONSOLE_USB_SERIAL_JTAG=y plus CONFIG_ESP_CONSOLE_SECONDARY_NONE=y;
• tracked-bundle SHA-256 manifest verifies and the app embeds 0.8.12 and RUVIEW_HELLO_OK_V1;
• workflow YAML parses and git diff --check passes.

The tracked binary will be reconciled once more against this PR's exact ESP-IDF 5.4 CI artifact before merge. Physical HELLO and configure verification remains an explicit post-build gate.

[proffesor-for-testing]

Exact-head verification for 2277b88:

• all GitHub checks are terminal with no failures; both ESP32-C6 builds and both ESP32-S3 size variants pass;
• local host delivery-contract tests and manifest verification pass;
• the CI-built C6 bundle identifies as 0.8.12; app size 1,078,224 bytes;
• the CI-built S3 4 MB app is 975,728 bytes and its manifest verifies.

Merge remains gated on the requested physical native USB/JTAG RUVIEW_HELLO_V1 check. Release assets intentionally publish only from a dedicated vX.Y.Z-esp32 tag; the unrelated server tag v2655 is not being retroactively populated.

[proffesor-for-testing]

Physical native USB/JTAG gate passed at exact head 2277b88 on 2026-09-11.

• identified /dev/cu.usbmodem14401 from ROM as ESP32-C6 revision 0.2 in USB-Serial/JTAG mode;
• flashed the exact CI artifact: bootloader, C6 partition table, initial OTA data, and 1,078,224-byte application;
• esptool verified the written hash for every region;
• the running firmware answered the nonce-bound command with RUVIEW_HELLO_OK_V1, chip=esp32c6, version=0.8.12, configured=1, and the matching nonce.

This closes the physical HELLO gate. The later RuView Mac qualification will separately exercise the complete in-app Scan -> Flash -> HELLO -> Configure -> server-verification journey.

[proffesor-for-testing]

2026-09-11 session handoff at exact head 2277b88a44873d0e152a604ee6d7c4231b08282e: all checks remain terminal green. The direct physical native USB/JTAG C6 gate passed, and mobile PR #27 has since merged the official-variant filename correction used by the Mac picker. A new Release Catalyst build from mobile main 0c188f5b is running as one UI PID plus one same-bundle helper.

Not yet claimed: the complete new-build in-app flow on all three C6 devices, any physical S3 result, or final three-C6 calibration/output evidence. One C6 carries the exact 0.8.12 CI image; the other two remain for in-app flashing. No S3 is attached. PR #1904 remains open for the next session decision.

[proffesor-for-testing]

Rebased onto main (45fc41ac). Head is now 0c1efda4; was DIRTY, now mergeable.

One commit dropped as redundant. 9907f062 fix(ci): size-gate each firmware variant against its own OTA slot loosened all three gates:

             was     this PR proposed
8mb         1152  ->  1638
4mb         1152  ->  1484
c6-4mb      1152  ->  1484

Since then af2d6981 landed on main and made a narrower change — raising only the 8 MB gate to 1200 KiB, with the reasoning recorded inline ("Current 8 MB image is 1163 KiB after USB onboarding. The OTA app partition is 2048 KiB"), and leaving 4 MB and C6 at 1152.

Main's values already cover this PR's own artifacts, so the commit had nothing left to do:

C6 onboarding app   1,078,224 bytes = 1052 KiB   vs c6-4mb gate 1152  OK
S3 8 MB CI artifact 1,191,440 bytes = 1163 KiB   vs 8mb gate    1200  OK

Taking main's tighter bound rather than re-applying the looser one — it is the newer deliberate decision and it is sufficient here. The firmware delivery commits (b1be9147, 0c1efda4) rebased cleanly.

Artifacts verified unchanged after the rebase, against the tracked SHA256SUMS.txt:

bootloader.bin         19,456 B   12e9fe5bd0533c3e…  OK
partition-table.bin     3,072 B   4c2cc4ffd52641e2…  OK
esp32-csi-node.bin  1,078,224 B   baff3940ede0ce16…  OK
ota_data_initial.bin    8,192 B                      OK

That application binary is the one flashed and verified on three physical ESP32-C6 boards from the Mac app — every region MD5-checked on readback, and the boards kept their node IDs and configuration through the flash.

[ruvnet]

Physical check on an ESP32-C6 behind a USB-UART bridge. Result: a regression for UART-bridge C6 boards. Please don't merge as-is.

Board: ESP32-C6FH4 rev v0.2 on a CP210x bridge wired to UART0, the common DevKitC-style layout. I built this PR at 0c1efda for esp32c6 with local ESP-IDF v5.4.

• The generated sdkconfig has CONFIG_ESP_CONSOLE_USB_SERIAL_JTAG=y and CONFIG_ESP_CONSOLE_SECONDARY_NONE=y.
• After reset, UART0 shows only the 9 ROM boot lines (entry 0x4086c110), then nothing. No app log, no serial_onboard: ... ready.
• RUVIEW_HELLO_V1 <nonce> over the bridge gets no reply. Onboarding through the UART port is impossible.
• With main @ dd02efe on the same board and cable, the full boot log appears on UART0 and HELLO returns RUVIEW_HELLO_OK_V1 ... chip=esp32c6.

This fixes native-USB C6 boards by breaking UART-bridge ones. ESP-IDF secondary consoles are output-only, as the PR notes. So either:

• ship two C6 variants (UART-primary default plus a USB-Serial/JTAG overlay), or
• keep UART0 primary and read onboarding input from both ports explicitly: install the usb_serial_jtag driver and poll usb_serial_jtag_read_bytes() alongside UART0, rather than relying on stdin.

The RuVector ESP32 decision firmware uses the first option (-DRD_USB_SERIAL_JTAG=ON). Happy to re-test on this board once there's an update.