map.average_precision killed when using large sample size

Hi,

I was using copairs to get a Phenotypic activity assesment based on mAP.

I was doing this using cpg0014 extracted features averaged per well. Below I show the shape of the data
```
>>> feats_meta[1].shape
(9216, 23)
```

I believe the issue I am experiencing could be solved by adding more resources to my VM but you might want to take care of an scenario when compute resources are limited but still desired to complete the job.

Memory available
```
total used free shared buff/cache available
Mem: 14Gi 1.7Gi 12Gi 1.0Mi 300Mi 12Gi
Swap: 0B 0B 0B
```

CPU info
```
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 46 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 4
On-line CPU(s) list: 0-3
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) CPU @ 2.30GHz
CPU family: 6
Model: 63
Thread(s) per core: 2
Core(s) per socket: 2
Socket(s): 1
Stepping: 0
BogoMIPS: 4599.99
```

Running the code below, it gets killed when trying to do map.average_precision on the whole dataset. Subsampling solves the issue and the job completes

```

from copairs import map
from copairs.matching import assign_reference_index

df_metadata = feats_meta[1]
feats = feats_meta[0]

reference_col = "Metadata_reference_index"
df_metadata_activity = assign_reference_index(
df_metadata,
"Metadata_broad_id == 'None'", # condition to get reference profiles (neg controls)
reference_col=reference_col,
default_value=-1,
)

# positive pairs are replicates of the same treatment
pos_sameby = ["Metadata_broad_id", reference_col]
pos_diffby = []
neg_sameby = []
# negative pairs are replicates of different treatments
neg_diffby = ["Metadata_broad_id", reference_col]

metadata = df_metadata_activity
profiles = feats.values

activity_ap = map.average_precision(
metadata, profiles, pos_sameby, pos_diffby, neg_sameby, neg_diffby
)

activity_ap = activity_ap.query("Metadata_broad_id != 'None'") # remove DMSO
activity_ap.to_csv("output/mAP/mAP.csv", index=False)

activity_map = map.mean_average_precision(
activity_ap, pos_sameby, null_size=1000000, threshold=0.05, seed=0
)
```

Start at the copairs.map.average_precision entry point and reproduce the failure with the reported 9,216-profile, 23-feature dataset, comparing it with the subsampled run. Trace where memory grows and define done as completing average_precision on the full dataset under constrained resources without being killed; the payload names no source files or tests.

Written by the indexing model from the issue text.