[QEC] Decoding server: guard tx-slot writes, drop dead code from #769

[bmhowe23]

Post-merge cleanups on the HOST_CALL request path:

• ResultWriter: split out payload_capacity(), which checks the slot against sizeof(RPCResponse) BEFORE subtracting. The old payload() evaluated capacity_ - sizeof(RPCResponse) first and relied on a later ||-term to catch the underflow.
• handle_get_corrections: take the payload pointer and capacity from the ResultWriter instead of recomputing them. The hand-rolled slot_size - sizeof(RPCResponse) bypassed the very guard the writer exists to provide; a slot too short for the response header now yields nullptr -> INTERNAL_ERROR rather than a wrapped capacity.
• Assert sizeof(RPCResponse) <= sizeof(RPCHeader). Every write_response caller gates on slot_size >= sizeof(RPCHeader) and then writes an RPCResponse into that slot; the structs are equal-sized today but live in the cudaq realtime headers and move independently of this repo.
• Drop DecodingServer::print_session_stats(), orphaned when the stats loop was moved into cudaqx_qec_decoding_server_print_stats(). Two copies of the QEC_DECODING_SERVER_DECODER_STATS line the test harness greps for could drift apart.
• Retire two CMake comments describing the removed worker threads.

New test RpcSlotWrite.ResultWriterRefusesASlotTooSmallForTheResponseHeader covers the exact-fit, undersized, null, and roomy slots.