[moretti]

Split out of #930 (1 of 4).

Summary

With DM_NVN_TSV set (off by default), a sampler records each frame's GPU clock, its limits and the process role in the frame's timing rows. A report script turns one session into a per-second timeline that says why each drop below the cap happened. That is how the other three PRs split out of #930 were found and checked.

dm/host/sysprof.m     every 100 ms, no root (IOReport through dlopen; Mac only)
  @gpu_mhz @gpu_on    GPU clock, share of time powered (powered is not busy)
  @gpu_w              GPU power
  @gpu_boost          clock state the boost controller asked for
  @gpu_cltm @gpu_droop  share of time a thermal / voltage-droop limit applied
  @role @thermal @low_power   process role (a change is logged), thermal state, Low Power Mode
  @ab                 with DM_AB: the switches on in the frame's phase
dm/tools/perf_report.py   one session -> timeline per second, a verdict per drop
                          (cpu, gpu-clock, gpu-limit, gpu, role, pacing); with DM_AB compares
                          the switch sets over the same session

Also: @late_ms no longer underflows when a display-link tick returns before the texture's slot time.

Documented in docs/development.md ("Why a frame rate fell: one session's profile").

Evidence

• Used on Breath of the Wild 1.9.0, MacBook Pro M5 Pro (Mac17,8), macOS 27.2, fullscreen, over runs 4–11 of Mac: steady 60 fps in fullscreen on ProMotion screens (Breath of the Wild) #930. It's what showed the GPU clock at its floor during failed probes (Frame step: a probe's grace lasts while its frames keep getting cheaper #1042) and the MetalFX present inside the frame (Present on the Mac: MetalFX scales at the snapshot, and pacing leaves the display link when the screen shifts under it #1045).
• Builds and tests on its own on current main: host and CPU gfx tests, sanitizers, tool tests, clang-tidy and ruff/pyright pass locally.

Merge Danger

Door: two-way. It does nothing unless DM_NVN_TSV is set.

Blast Radius: diagnostics.

The weak dm_sysprof_* stubs in hle_nvn.c keep the graphics unit tests linking without sysprof.m.

regress.py impact: Breath of the Wild, Pokémon Scarlet and Tears of the Kingdom reach the changed present code (the rows are only written with DM_NVN_TSV). Hades is reached through the host Makefile rule. None of them is on this machine, so regress.py check --games was not run.