Issues with `Timer` `start` and `stop` methods when using `collect_performance: true` config option

[fmalatino]

Describe the bug
When setting the configuration option collect_performance: true for dace:cpu runs of more than one time step, the start and stop methods of the TImer class in ndsl/performance/timer.py throw the following exceptions:

2025-11-19 13:16:39|INFO|rank 0|ndsl.logging:CubedToLatLon
Exception ignored on calling ctypes callback function: <function flatten_callback.<locals>.make_cb.<locals>.cb_func at 0x7f14739942c0>
Traceback (most recent call last):
  File "/ncrc/home2/Frank.Malatino/.conda/envs/throw2/lib/python3.11/site-packages/dace/frontend/python/preprocessing.py", line 441, in cb_func
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/ncrc/home2/Frank.Malatino/.conda/envs/throw2/lib/python3.11/site-packages/ndsl/performance/timer.py", line 87, in __exit__
    self.timer.stop(name)
  File "/ncrc/home2/Frank.Malatino/.conda/envs/throw2/lib/python3.11/site-packages/ndsl/performance/timer.py", line 43, in stop
    self._accumulated_time[name] += time() - self._clock_starts.pop(name)
                                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
KeyError: 'mainloop'
2025-11-19 13:16:39|INFO|rank 0|ndsl.logging:Finished stepping 0
/ncrc/home2/Frank.Malatino/.conda/envs/throw2/lib/python3.11/site-packages/ndsl/performance/collector.py:75: RuntimeWarning: Retrieved times while clocks are still going, incomplete times are not included: ['DynCore']
  self.times_per_step.append(self.timestep_timer.times)
/ncrc/home2/Frank.Malatino/.conda/envs/throw2/lib/python3.11/site-packages/ndsl/performance/collector.py:76: RuntimeWarning: Retrieved hit counts while clocks are still going, incomplete times are not included: ['DynCore']
  self.hits_per_step.append(self.timestep_timer.hits)
2025-11-19 13:16:39|INFO|rank 0|ndsl.logging:diagnostics for step 2000-01-01 00:03:45 started
2025-11-19 13:16:39|INFO|rank 0|ndsl.logging:diagnostics for step 2000-01-01 00:03:45 finished
2025-11-19 13:16:39|INFO|rank 0|ndsl.logging:DynCore
2025-11-19 13:16:39|INFO|rank 0|ndsl.logging:TracerAdvection
2025-11-19 13:16:39|INFO|rank 0|ndsl.logging:DynCore
Exception ignored on calling ctypes callback function: <bound method Timer.clock.<locals>.Wrapper.__enter__ of <ndsl.performance.timer.Timer.clock.<locals>.Wrapper object at 0x7f14737072d0>>
Traceback (most recent call last):
  File "/ncrc/home2/Frank.Malatino/.conda/envs/throw2/lib/python3.11/site-packages/ndsl/performance/timer.py", line 83, in __enter__
    self.timer.start(name)
  File "/ncrc/home2/Frank.Malatino/.conda/envs/throw2/lib/python3.11/site-packages/ndsl/performance/timer.py", line 28, in start
    raise ValueError(f"clock already started for '{name}'")
ValueError: clock already started for 'DynCore'
2025-11-19 13:16:39|INFO|rank 0|ndsl.logging:TracerAdvection

To Reproduce
Run a configuration with backend: dace:cpu and collect_performance: true

Expected behavior
The timer should be able to handle that the clock has already started for the dynamical core, and that mainloop exists as a key for the _accumulated_time and _clock_starts dictionaries.

System Environment
Describe the system environment, include:

• OS: SUSE Linux Enterprise Server 15 SP6
• Backend used: dace:cpu
• Environment variables set: FV3_DACEMODE=Build
• Compiler(s): Type and version: gcc-native/12.3
• MPI type, and version: cray-mpich-abi/8.1.28
• netCDF Version: N/A
• Configure options: collect_performance: true
• If this bug came from a model run, which model: baroclinic_c12.yaml in Pace, with a 1x1 layout, and total runtime of 15 minutes

[romanc]

Frank, can you share NDSL-only reproduction steps without having to run a model? The timer class has a bunch of unit tests and the following minimal reproducer works just fine on my machine

import time

from ndsl.performance import Timer

timer = Timer()

for i in range(5):
    with timer.clock("sleep"):
        time.sleep(0.2)

print(f"Accumulated timings: {timer.times}")