Optimize pwndbg.commands.ai import time (#1984)

Before:

```
root@pwndbg:~/pwndbg# PYTHONPROFILEIMPORTTIME=1 gdb --batch 2>&1 | grep 'pwndbg.commands.ai'
import time:      5363 |     174096 |   pwndbg.commands.ai
root@pwndbg:~/pwndbg# PYTHONPROFILEIMPORTTIME=1 gdb --batch 2>&1 | grep 'pwndbg.commands.ai'
import time:      1885 |     154032 |   pwndbg.commands.ai
root@pwndbg:~/pwndbg# PYTHONPROFILEIMPORTTIME=1 gdb --batch 2>&1 | grep 'pwndbg.commands.ai'
import time:      2085 |     148208 |   pwndbg.commands.ai
root@pwndbg:~/pwndbg# PYTHONPROFILEIMPORTTIME=1 gdb --batch 2>&1 | grep 'pwndbg.commands.ai'
import time:      1903 |     146919 |   pwndbg.commands.ai
```

After:

```
root@pwndbg:~/pwndbg# PYTHONPROFILEIMPORTTIME=1 gdb --batch 2>&1 | grep 'pwndbg.commands.ai'
import time:      5522 |     105163 |   pwndbg.commands.ai
root@pwndbg:~/pwndbg# PYTHONPROFILEIMPORTTIME=1 gdb --batch 2>&1 | grep 'pwndbg.commands.ai'
import time:      1842 |      88943 |   pwndbg.commands.ai
root@pwndbg:~/pwndbg# PYTHONPROFILEIMPORTTIME=1 gdb --batch 2>&1 | grep 'pwndbg.commands.ai'
import time:      1780 |      85127 |   pwndbg.commands.ai
root@pwndbg:~/pwndbg# PYTHONPROFILEIMPORTTIME=1 gdb --batch 2>&1 | grep 'pwndbg.commands.ai'
import time:      1825 |      83504 |   pwndbg.commands.ai
root@pwndbg:~/pwndbg# PYTHONPROFILEIMPORTTIME=1 gdb --batch 2>&1 | grep 'pwndbg.commands.ai'
import time:      1806 |      87045 |   pwndbg.commands.ai
root@pwndbg:~/pwndbg# PYTHONPROFILEIMPORTTIME=1 gdb --batch 2>&1 | grep 'pwndbg.commands.ai'
import time:      1756 |      81687 |   pwndbg.commands.ai
```

pwndbg/commands/ai.py

@@ -13,7 +13,6 @@ import pprint
 import re
 
 import gdb
-import requests
 
 import pwndbg
 import pwndbg.color.message as M
@@ -70,6 +69,13 @@ dummy = False
 verbosity = 0
 
 
+def _requests():
+    """Lazy import requests since its import is quite heavy"""
+    import requests
+
+    return requests
+
+
 def set_dummy_mode(d=True) -> None:
     global dummy
     dummy = d
@@ -275,11 +281,11 @@ def query_openai_chat(prompt, model="gpt-3.5-turbo", max_tokens=100, temperature
         "temperature": temperature,
     }
     url = "https://api.openai.com/v1/chat/completions"
-    r = requests.post(
+    r = _requests().post(
         url,
         data=json.dumps(data),
         headers={"Content-Type": "application/json"},
-        auth=("Bearer", config.ai_openai_api_key),  # type: ignore[arg-type]
+        auth=("Bearer", config.ai_openai_api_key),
     )
     res = r.json()
     if verbosity > 0:
@@ -315,11 +321,11 @@ def query_openai_completions(prompt, model="text-davinci-003", max_tokens=100, t
         "stop": ["\n\nHuman:"],
     }
     url = "https://api.openai.com/v1/completions"
-    r = requests.post(
+    r = _requests().post(
         url,
         data=json.dumps(data),
         headers={"Content-Type": "application/json"},
-        auth=("Bearer", config.ai_openai_api_key),  # type: ignore[arg-type]
+        auth=("Bearer", config.ai_openai_api_key),
     )
     res = r.json()
     if verbosity > 0:
@@ -367,7 +373,7 @@ def query_anthropic(prompt, model="claude-v1", max_tokens=100, temperature=0.0):
     }
     headers = {"x-api-key": config.ai_anthropic_api_key.value, "Content-Type": "application/json"}
     url = "https://api.anthropic.com/v1/complete"
-    response = requests.post(url, data=json.dumps(data), headers=headers)
+    response = _requests().post(url, data=json.dumps(data), headers=headers)
     data = response.json()
     try:
         return data["completion"].strip()
@@ -378,7 +384,7 @@ def query_anthropic(prompt, model="claude-v1", max_tokens=100, temperature=0.0):
 
 def get_openai_models():
     url = "https://api.openai.com/v1/models"
-    r = requests.get(url, auth=("Bearer", config.ai_openai_api_key))  # type: ignore[arg-type]
+    r = _requests().get(url, auth=("Bearer", config.ai_openai_api_key))
     res = r.json()
     if verbosity > 0:
         print(M.warn(pprint.pformat(res)))