fix(build): move boot_log into eboot_core so the core library links standalone

[Kartikey1306]

Summary

eboot_core cannot link on its own. core/recovery.c, core/boot_policy.c and
stage0/jump_stage1.c all call eos_boot_log_*(), but boot_log.c was
compiled into eboot_stage1 — which itself links PUBLIC against eboot_core.

That circular dependency breaks two targets:

• test_recovery links only eboot_core and fails to build.
• ebldr_stage0 compiles stage0/jump_stage1.c, which calls
  eos_boot_log_append() three times, and links
  eboot_core board_${EBLDR_BOARD} without eboot_stage1. It cannot link
  for any board.

The fix is to put the code in the layer that uses it.

The bug

eboot_stage1  ──links──▶  eboot_core
      ▲                        │
      └────── needs ───────────┘   core/recovery.c, core/boot_policy.c,
             (boot_log.c)          stage0/jump_stage1.c

On master:

$ cmake --build build -j
Undefined symbols for architecture arm64:
  "_eos_boot_log_append", referenced from:
      _eos_recovery_enter in libeboot_core.a[6](recovery.c.o)
      ...
  "_eos_boot_log_get_head", referenced from:
      _recovery_collect_boot_log_entries in libeboot_core.a[6](recovery.c.o)
  "_eos_boot_log_read", referenced from:
      _recovery_collect_boot_log_entries in libeboot_core.a[6](recovery.c.o)
ld: symbol(s) not found

The build exits 2 and ctest reports 13/14 with test_recovery Not Run.

The stage0 half is the part that matters

test_recovery is a test. ebldr_stage0 is firmware. It has exactly the
same unmet dependency, and it is invisible in CI because EBLDR_BOARD defaults
to "none", so the target is never created in a host build
(CMakeLists.txt:151). It would only surface on a real cross-compile.

Demonstrated by linking stage0's requirement against each version of the
library:

$ cat stage0_sim.c
extern void eos_boot_log_append(uint32_t event, uint32_t slot, uint32_t detail);
int main(void) { eos_boot_log_append(1, 0, 0); return 0; }

# against eboot_core on master
Undefined symbols: "_eos_boot_log_append"     ld: symbol(s) not found

# against eboot_core with this change
linked OK / ran OK

Approach

Move boot_log.c from stage1/ to core/.

It is core-layer code already: it includes only eos_types.h, eos_hal.h and
string.h, and implements a flash-backed circular log with no stage-1
dependency of any kind. Its callers are split across core, stage0 and stage1 —
the one layer it does not belong to is stage1.

The file is moved verbatim. Git records a pure rename and the SHA-256 is
identical before and after (b345d030…de99); the only edits are the two CMake
source-list lines.

Why not link eboot_stage1 into the test

That is what #35 does, and it does fix test_recovery. I went the other way for
two reasons:

1. It leaves eboot_core non-self-contained, so ebldr_stage0 stays
   broken — the test passes but the firmware target still cannot link.
2. It links a stage-1 library into a core-level unit test, which inverts the
   layering in the test as well.

Because this fixes the root cause, tests/CMakeLists.txt needs no change at
all, so there is no textual conflict with #35 — whichever lands first, the
other is either redundant or applies cleanly. Happy to close this in favour of
#35 if maintainers prefer the narrower fix, but the stage0 break would then want
its own issue.

Testing

$ cmake -S . -B build -DEBLDR_BUILD_TESTS=ON -DCMAKE_BUILD_TYPE=Debug
$ cmake --build build -j
BUILD=0                       # master exits 2

$ ctest --test-dir build
100% tests passed out of 14   # master: 13/14, test_recovery Not Run

test_recovery builds and passes. No other test changes behaviour.

I specifically checked for a duplicate-symbol clash: test_boot_log links
eboot_core and defines its own eos_boot_log_* stubs. It still builds and
passes, because the linker only pulls boot_log.c.o from the archive to resolve
an undefined symbol, and that test defines them all itself.

Two findings I am reporting rather than fixing

Both turned up while tracing this and are out of scope for a layering fix:

1. include/eos_boot_log.h does not describe the implementation. It
   declares a different, incompatible API under the same names:

   	header	boot_log.c
   eos_boot_log_init	int (void)	void (uint32_t head)
   eos_boot_log_read	int (entry_t *, uint32_t)	int (uint32_t, entry_t *)
   eos_boot_log_get_head	absent	uint32_t (void)
   count / flush / get_latest / event_name	declared	not implemented

   Nothing breaks today only because no non-test source includes the header —
   every caller writes its own local extern declaration. Anyone who does
   include it gets eos_boot_log_read(entries, max_count) compiled against an
   implementation whose first parameter is an index, i.e. a pointer
   reinterpreted as uint32_t.

2. tests/unit/test_boot_log.c tests its own stubs. It includes
   eos_boot_log.h, then defines every function in that header itself, and
   asserts against those definitions. It exercises no line of boot_log.c.

Reconciling the header and rewriting that test against the real implementation
(it needs a simulated flash HAL, like test_image_verify.c has) is a
self-contained follow-up. I did not want to bundle a rewrite of a passing test
into a two-line build fix. Happy to open an issue or a follow-up PR.

Limitations and considerations

• No behaviour change. Same code, same symbols, different archive. Nothing
  is added to or removed from any firmware image; eboot_stage1 still gets
  boot_log transitively through eboot_core.
• Cross-compile not exercised here. I do not have arm-none-eabi-gcc on
  this machine, so ebldr_stage0 is NOT RUN — the break and the fix are
  demonstrated by reproducing stage0's link requirement natively (above) rather
  than by building the firmware target.
• Line endings preserved. CMakeLists.txt is CRLF and CHANGELOG.md is
  mixed; edits were applied byte-wise so untouched lines keep their endings.

Type of Change

• fix — Bug fix
• build — Build system changes

Changes

• stage1/boot_log.c → core/boot_log.c (verbatim move).
• CMakeLists.txt — source listed under eboot_core instead of eboot_stage1.
• CHANGELOG.md — Unreleased entry.

Pre-Submission Checklist

• Compiles without warnings
• All tests pass — 14/14, up from 13/14
• Fixes a test that could not previously build
• No behaviour change; file moved verbatim (identical SHA-256)
• No conflict with fix: link eboot_stage1 into test_recovery #35