E27: Reentrancy review and hardening

[DavidCozens]

Motivation

The logging path crosses task boundaries on RTOS: application tasks call SolidSyslog_Log, a service task calls SolidSyslog_Service, and integrator-supplied callbacks (string functions, clock, error handler, onSwitch) can fire from yet other tasks. Some reentrancy is handled by construction (per-call SolidSyslogFormatter on the caller's stack; SolidSyslogAtomicCounter for sequenceId; mutex-protected SolidSyslogBuffer). Some is documented contract (single global error handler slot). And some is actually broken — see the first child story.

This epic exists to:

• Audit the current state once, end to end.
• Document which parts of the API are reentrant by construction, which require integrator contract, and which have actual bugs.
• Fix the bugs.
• Tighten the contract where it's currently implicit.

Likely children (refine when we pick this up)

• OriginSd shared scratch fix — first child, filed as a story under this epic.
• Reentrancy audit doc — enumerate every object in the logging path, classify each.
• SolidSyslogSwitchingSender currentIndex race — UART task writes while Service task reads; plain int, tear-free on most platforms but UB by the C standard.
• Callback contract clarity — the integrator's reentrancy obligations for string/clock/error/onSwitch callbacks are currently implicit. Make them explicit in the public-header audience table in CLAUDE.md and in each header.

Possibly in scope, possibly its own thing

• Whether the SD pattern itself should be restructured (base class / helper) — that decision sits on the OriginSd bug story.
• Whether the error handler slot should grow synchronisation, or stay documented-as-not-synchronised.
• Hot-swap config during send → currently slated for the future Reconfiguration epic; revisit the boundary.

Out of scope

• The long-running multi-task concurrent-Log test application — belongs alongside fuzzing in E25: Fuzz testing and adversarial coverage #256 (E25).
• New transport reentrancy (covered when each transport is added).

Open questions

• Audit-only first slice, then act on findings — or interleave audit with fixes as they're found?
• Are integrator-callback contract changes API-breaking enough to warrant a feat! bump, or can they go as docs?
• Does the SD restructure (if Option Y on the OriginSd story is chosen) live here, or under E07: Structured Data #9 (E07 Structured Data)?
• Should the concurrent-Log test framework live here or in E25: Fuzz testing and adversarial coverage #256 (E25 Fuzz testing)? Leaning E25.

Status

Not started. Will be picked up after the memory-consumption sequence (CMake-sizing epic, static-create discussion). Exact slot TBD. Refine as we get closer.