Align DMA memcpy with the EF's Accelerated Memory Operations standard

[nicole-graus]

Motivation

Aligns the DMA memcpy accelerator from #874 with the EF's "Accelerated Memory Operations" proposal (eth-act/zkevm-standards#32), which requires a vendor to pin down how its memcpy wins symbol resolution, to document which mechanism it relies on, and encourages reporting what the acceleration costs.

Description

• Defines memcpy in the always-linked entrypoint object, next to _start, so it wins resolution regardless of codegen-unit partitioning — mechanism (1) of the standard, documented in docs/general_flow.md.
• Adds a guest whose copies are only compiler-emitted, so a silent fallback to the weak compiler-builtins definition fails a test instead of quietly costing performance.
• Reports Dma bytes and Dma rows alongside Dma calls in execute --cycles, since one memcpy becomes several ecalls and the call count alone is a poor cost proxy.

[github-actions]

Benchmark Results for modified programs 🚀

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
head hashmap	91.8 ± 18.5	84.9	144.5	1.00