Adopt m68kemu v1.5.0 IRQ source and simplify device/config internals

[jenska]

Finishes the v1.5.0 migration and the cleanup it opens up.

m68kemu v1.5.0 adoption:

• Build the CPU with WithDeferredReset and run Machine.Reset once during construction, making it the single reset path. This also fixes a latent bug: the ROM fast-memory window was only installed from Machine.Reset, which the desktop and headless runners never call, so the shipped binary had been running entirely on the bus path.
• Route device interrupts through a single level-sensitive IRQSource (machineIRQ over GLUE/MFP/FDC) that the core samples each instruction, deleting dispatchInterrupts, maskedAutovectorPulse, the irqSources slice and the InterruptSource interface. GLUE now drives a held HBL/VBL autovector line instead of queuing pulses; a blank interrupt the CPU is masking is taken shortly after the mask clears rather than dropped.
• EnableTrace disassembly comes from TraceInfo.Mnemonic; drop the direct m68kdasm dependency (m68kemu still pulls it in).

Internal cleanup, no behaviour change:

• PAL/NTSC raster constants and the FrameHz>=55 rule moved to config.Config.Video(); GLUE and MFP timing setup collapse to a few lines.
• GLUE dropped its unused system-control register; $FF8006 is now a shared devices.ScratchRegion and GLUE is no longer bus-mapped.
• DrainInterrupts removed from the devices in favour of PendingIRQ/AckIRQ plus a drainIRQ test helper.
• The JSON config loader is table-driven (config.Config.jsonFields).
• Removed the dead Machine.cartridge field.

The CycleScheduler adoption (proposal 3.2) was re-attempted against v1.5.0 and reverted again: correct, but a 120-frame EmuTOS boot regressed ~20% because the devices advance per-quantum, not event-driven. Recorded in docs/m68kemu-api-proposal.md.