Object system used in Aether library

AE_OBJECTS_RAW_TELE=ON connects the tagged internal AE_LOG_MACRO_TAGGED checkpoints to the aether-tele raw runtime. It adds a dependency on aether::tele; local development can supply it with -DCPM_aether-tele_SOURCE=/path/to/aether-tele.

Object telemetry uses its own "objects" space and stable explicit callsite indices. Its public sink is declared in aether-objects/telemetry.h:

auto& objects = ae::objects_tele::ObjectTeleSink::Instance();
objects.SetBackend(backend);  // shared_ptr<ae::tele::raw::IBackend>
objects.BindWriter(0);       // stable ID for this owner thread, before logging
// Run object/domain operations.
bool saved = objects.DrainCurrent();

The runtime mask independently controls logs, console, counts, timing and fields for this space. The existing debug boundary is retained: NDEBUG or AE_NO_DEBUG_LOG=1 removes internal log calls, including argument evaluation. A custom AE_LOG_MACRO supplied through AE_LOGGER_INCLUDE retains precedence. The standalone default leaves raw telemetry disabled and keeps the old console logger. No object IDs, stored object data or pointer ownership rules change.

Each thread owns its writer. DrainAll() serializes each checkpoint with its writer and includes buffered data left by exited threads. A shared backend must serialize complete blocks and implement all-or-retry acknowledgement. Bind stable writer IDs to restore their statistics across restarts. When replacing a backend, use SetBackend's migration callback to transfer its accepted history before the new binding is published. Do not recursively save the same domain from its telemetry drain callback; telemetry storage and domain operations must have independent lifetimes and IO paths.

The domain calls IDomainStorageWriter::Commit() after a version serializes successfully, before notifying Saved(). Its default implementation preserves custom/RAM writers. Direct users of the writer API must call Commit() when using filesystem storage.

FileSystemStdStorage writes to an exclusively created temporary sibling and publishes it with an atomic replacement only after flushing the complete file. Abandoning a writer or failing serialization removes its own temporary file and keeps the previous committed version. A crash can leave a temporary sibling; ordinary readers ignore it and another writer never removes it. POSIX also synchronizes the parent directory after replacement. If that final sync fails, commit reports an error; the visible file is still a complete version, but its durability acknowledgement is uncertain. Each version file is a transaction; this does not make a whole multi-object graph one transaction.