[MasterLaplace]

What

cpu_topology_get_logical_slot gives each processor the slot its own APIC ID register names. It read one identifier recorded on the bootstrap processor, so every processor got slot 0 and the heap's first domain.

• Before the local APICs run (apic_is_running), only the bootstrap processor runs, and it keeps the recorded identifier.
• An identifier without a slot gives CPU_TOPOLOGY_NO_SLOT instead of being registered.
• The docs of cpu_topology_get_local_apic_id and its setter now say the bootstrap processor's.

Two heap tests relied on the bug: each application processor's start-up blocks landed in domain 0, leaving domain 1 empty. They now empty domain 1 first.

Verify

• Red on main: the_running_processor_keeps_the_slot_of_its_own_register and every_processor_finds_its_own_slot.
• Server: q35 861/0, -cpu max 863/0, -smp 4 861/0, pc 855/0, -smp 1 856/0. Satellite: q35 711/0, pc -cpu max 706/0.
• 3 of 4 mutants fail a test. apic_is_running always true survives: on QEMU the early read names no slot, which callers treat as slot 0.
• Every budget within tolerance; server tick +0.03 %.

Not in it

• The SRAT binds only the bootstrap processor: The SRAT binds only the bootstrap processor to its NUMA node: it is read before the other processors have slots #802.
• A borrowed heap block returns to its owner, so an empty domain keeps borrowing (Per-CPU then per-NUMA allocation domains #125).

Closes #792