Add support for exl2 quantization

Mostly straightforward, changes to existing code:

* Wrap quantizer parameters in a small wrapper to avoid passing
  around untyped tuples and needing to repack them as a dict.
* Move scratch space computation to warmup, because we need the
  maximum input sequence length to avoid allocating huge
  scratch buffers that OOM.

integration-tests/models/__snapshots__/test_flash_gemma_exl2/test_flash_gemma_exl2.json

@@ -0,0 +1,89 @@
+{
+  "details": {
+    "best_of_sequences": null,
+    "finish_reason": "length",
+    "generated_tokens": 10,
+    "prefill": [
+      {
+        "id": 2,
+        "logprob": null,
+        "text": "<bos>"
+      },
+      {
+        "id": 2015,
+        "logprob": -672.5,
+        "text": "Test"
+      },
+      {
+        "id": 3853,
+        "logprob": -16.65625,
+        "text": " request"
+      }
+    ],
+    "seed": null,
+    "tokens": [
+      {
+        "id": 604,
+        "logprob": -1.0458984,
+        "special": false,
+        "text": " for"
+      },
+      {
+        "id": 10665,
+        "logprob": -0.7919922,
+        "special": false,
+        "text": " API"
+      },
+      {
+        "id": 37875,
+        "logprob": -0.76171875,
+        "special": false,
+        "text": " endpoint"
+      },
+      {
+        "id": 235292,
+        "logprob": -1.0478516,
+        "special": false,
+        "text": ":"
+      },
+      {
+        "id": 109,
+        "logprob": -0.81152344,
+        "special": false,
+        "text": "\n\n"
+      },
+      {
+        "id": 688,
+        "logprob": -0.0390625,
+        "special": false,
+        "text": "**"
+      },
+      {
+        "id": 35447,
+        "logprob": -0.2232666,
+        "special": false,
+        "text": "Endpoint"
+      },
+      {
+        "id": 66058,
+        "logprob": -0.00017559528,
+        "special": false,
+        "text": ":**"
+      },
+      {
+        "id": 102001,
+        "logprob": -0.11376953,
+        "special": false,
+        "text": " `/"
+      },
+      {
+        "id": 3564,
+        "logprob": -0.25610352,
+        "special": false,
+        "text": "api"
+      }
+    ],
+    "top_tokens": null
+  },
+  "generated_text": " for API endpoint:\n\n**Endpoint:** `/api"
+}

integration-tests/models/__snapshots__/test_flash_gemma_exl2/test_flash_gemma_exl2_all_params.json

@@ -0,0 +1,89 @@
+{
+  "details": {
+    "best_of_sequences": null,
+    "finish_reason": "length",
+    "generated_tokens": 10,
+    "prefill": [
+      {
+        "id": 2,
+        "logprob": null,
+        "text": "<bos>"
+      },
+      {
+        "id": 2015,
+        "logprob": -672.0,
+        "text": "Test"
+      },
+      {
+        "id": 3853,
+        "logprob": -16.609375,
+        "text": " request"
+      }
+    ],
+    "seed": 0,
+    "tokens": [
+      {
+        "id": 604,
+        "logprob": 0.0,
+        "special": false,
+        "text": " for"
+      },
+      {
+        "id": 10665,
+        "logprob": -0.16027832,
+        "special": false,
+        "text": " API"
+      },
+      {
+        "id": 53785,
+        "logprob": 0.0,
+        "special": false,
+        "text": " Gateway"
+      },
+      {
+        "id": 37875,
+        "logprob": -0.6328125,
+        "special": false,
+        "text": " endpoint"
+      },
+      {
+        "id": 235292,
+        "logprob": -0.20141602,
+        "special": false,
+        "text": ":"
+      },
+      {
+        "id": 109,
+        "logprob": 0.0,
+        "special": false,
+        "text": "\n\n"
+      },
+      {
+        "id": 688,
+        "logprob": -0.4741211,
+        "special": false,
+        "text": "**"
+      },
+      {
+        "id": 6217,
+        "logprob": -0.43286133,
+        "special": false,
+        "text": "URL"
+      },
+      {
+        "id": 66058,
+        "logprob": 0.0,
+        "special": false,
+        "text": ":**"
+      },
+      {
+        "id": 102001,
+        "logprob": 0.0,
+        "special": false,
+        "text": " `/"
+      }
+    ],
+    "top_tokens": null
+  },
+  "generated_text": "Test request for API Gateway endpoint:\n\n**URL:** `/"
+}