makes POST /activation/all support pagination

[anthonyduong9]

Problem

• For POST /activation/all, the request has a num_results property, but no offset property.
• There's a TODO to return a better error for when ActivationAllPostRequest.sort_by_token_indexes has out of range indexes.

Fix

• Adding an offset property in all.yaml.
• Running make setup-all-inference VERSION=1.4.0.
• Adding logic to use the offset and number of results, to slice results.
• Replacing 500 with 400 for when ActivationAllPostRequest.sort_by_token_indexes has out of range indexes.

Testing

From apps/inference, I ran poetry run python start.py, then

curl -X POST http://127.0.0.1:5002/v1/activation/all \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "The narrative of ... unattainable.",
    "model": "gpt2-small",
    "source_set": "res-jb",
    "selected_sources": ["1-res-jb"],
    "sort_by_token_indexes": [4],
    "ignore_bos": true,
    "num_results": 5,
    "offset": 0
  }' | jq '.activations[].index'

, then

curl -X POST http://127.0.0.1:5002/v1/activation/all \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "The narrative of ... unattainable.",
    "model": "gpt2-small",
    "source_set": "res-jb",
    "selected_sources": ["1-res-jb"],
    "sort_by_token_indexes": [4],
    "ignore_bos": true,
    "num_results": 5,
    "offset": 4
  }' | jq '.activations[].index'

and

curl -X POST http://127.0.0.1:5002/v1/activation/all \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Short prompt",
    "model": "gpt2-small",
    "source_set": "res-jb",
    "selected_sources": ["1-res-jb"],
    "sort_by_token_indexes": [10],
    "ignore_bos": true,
    "num_results": 5,
    "offset": 0
  }'

.

Fixes #74

Depends on #80