Kilosort4 capsule fails on import due to Numba cache locator error in si-0.103.2 container

Hello,

I have this issue using aind-ephys-pipeline with container tag si-0.103.2, where the Kilosort4 capsule fails immediately on import:
```
RuntimeError: cannot cache function '_find_duplicated_spikes_keep_last_iterative': no locator available for file '/home/miniconda3/lib/python3.11/site-packages/spikeinterface/curation/curation_tools.py'
```
Pipeline context:
- aind-ephys-pipeline revision: 9b4b4a2d88
- capsule: aind-ephys-spikesort-kilosort4
- capsule commit: dbe0c9683433f0d10057b9db800c62654ee1abf5
- container: ghcr.io/allenneuraldynamics/aind-ephys-spikesort-kilosort4:si-0.103.2
- Numba version in container: 0.60.0
- SpikeInterface path: /home/miniconda3/lib/python3.11/site-packages/spikeinterface

Failure occurs before sorting starts:
```
File "capsule/code/run_capsule.py", line 23, in
import spikeinterface.curation as sc
File "/home/miniconda3/lib/python3.11/site-packages/spikeinterface/curation/curation_tools.py", line 79, in
@numba.jit(nopython=True, nogil=True, cache=True)
...
RuntimeError: cannot cache function '_find_duplicated_spikes_keep_last_iterative': no locator available for file '/home/miniconda3/lib/python3.11/site-packages/spikeinterface/curation/curation_tools.py'
```
I tested setting NUMBA_CACHE_DIR to a writable path inside the container, but the same error occurs. Setting NUMBA_DISABLE_JIT=1 avoids the import error, but that is not really a viable strategy.

This looks like the container/package layout makes Numba unable to create a cache locator for SpikeInterface functions decorated with cache=True. Could the container or capsule be updated so this import works without disabling JIT globally?

Thanks!