zephyr-cp: do not rerun the sysbuild configure for every language build

[dhalbert]

Claude found and fixed this, reviewed by @dhalbert.

Problem

ports/zephyr-cp/Makefile gives west build cmake arguments on every invocation, and west reruns the full two-pass sysbuild configure whenever it gets any. The release build does that 17 times per board, once per language, although the language only affects the build_circuitpython.py step; TRANSLATION is not used anywhere else in cmake. On the 4-core runner that is an estimated 10-20 s of each language build after #11416, roughly 80-150 runner-minutes per main build across the 30 boards.

Changes

• build_circuitpython.py takes the translation from a TRANSLATION environment variable when set, then from the cmake argument, then en_US.
• The Makefile exports TRANSLATION to the recipe and records the west arguments (board, shields, cmake arguments) in <build>/west_configure_args.txt. It runs the configuring west build -b ... --sysbuild -- ... only when the build directory is new or the recorded arguments differ; otherwise it runs west build -d <build>, which is ninja only. ninja still reconfigures by itself when a Kconfig, overlay or CMake file changes, so the record only has to cover argument changes such as DEBUG=1 or a different SHIELD.
• The record is written before west runs and removed if west fails, so a failed configure cannot leave the directory configured with arguments the record does not show.

Together with #11416, this makes the zephyr-cp builds work like the make-based ports: building another language regenerates the translation data, recompiles the few files that use it and relinks, with no configure step.

Verification

Feather RP2040 build, locally, with #11422 and #11423 applied.

Step	Configure passes	CircuitPython compiles
fresh en_US	2	412
no change	0	0
de_DE	0	3
back to en_US	0	2
DEBUG=1	2, debug.conf merged, record updated	412 (flags changed)
DEBUG=1 again	0	0
without DEBUG	2	412
configure made to fail (missing fragment)	2, exits non-zero, record removed
next normal build	2, succeeds	0
and once more	0	0

The incremental de_DE binary is byte-identical to a clean de_DE build at the same path. cptools tests pass.

Depends on #11423 for DEBUG=1 to work at all; independent of it otherwise.

[mikeysklar]

Turning DEBUG=1 off again does not work on the 8 boards with no board.conf: make DEBUG=1 then plain make stays a debug build until make clean. Main does the same, so it is not from this PR. Passing the argument every time, even empty, fixed it for me, in place of the ifneq/endif around it:

WEST_CMAKE_ARGS += -Dzephyr-cp_EXTRA_CONF_FILE=$(CP_BOARD_CONF)

Step	main and this PR	with the change
fresh	normal	normal
DEBUG=1	debug	debug
no DEBUG	debug (wrong)	normal

Checked by elf md5 and .config on raspberrypi_rpi_pico_zephyr, Linux, make 4.3. The repeat build still skips the configure with the change (3 s).

The PR itself works as described here. The inline suggestions only shorten comments.

[dhalbert]

Turning DEBUG=1 off again does not work on the 8 boards with no board.conf: make DEBUG=1 then plain make stays a debug build until make clean. Main does the same, so it is not from this PR. Passing the argument every time, even empty, fixed it for me, in place of the ifneq/endif around it:

I did this in a new commit; could you re-check that? Thanks.

[dhalbert]

Claude wrote this:

Re-checked the branch at 027d6ff, which carries @mikeysklar's always-pass-the-fragment change (cce234f) and #11423.

Board	Step	Configure passes	debug.conf merged	recorded fragment
raspberrypi_rpi_pico_zephyr (no board.conf)	fresh	2	no	""
	again	0		""
	DEBUG=1	2	yes	debug.conf
	plain	2	no, .config back to normal	""
	plain again	0
	de_DE	0
adafruit_feather_rp2040_zephyr (has board.conf)	fresh	2	no	board.conf
	DEBUG=1	2	yes	board.conf;debug.conf
	plain	2	no	board.conf
	plain again	0

Dropping DEBUG=1 now reconfigures on both kinds of board, the semicolon quoting from #11423 survives the change, the empty argument draws no cmake complaint, and language switches still skip the configure.

[mikeysklar]

Re-checked 027d6ff on rpi_pico_zephyr. DEBUG=1 then plain make now goes back to the normal build, and the repeat build still skips the configure. Looks good, approved.