Simplify the attention function (#2609)

* Simplify the `attention` function

- Use one definition rather than multiple.
- Add `key`/`value` arguments, so that we don't need the
  `PREFILL_IN_KVCACHE` constant.
- Make it kwargs-only (to avoid mixing up the various `Tensor` args).

* Fixup flashinfer support

server/text_generation_server/layers/attention/__init__.py

@@ -8,23 +8,20 @@
     raise ImportError("`USE_FLASH_ATTENTION` is false.")
 if SYSTEM == "cuda":
     from .cuda import (
-        PREFILL_IN_KV_CACHE,
         SUPPORTS_WINDOWING,
         attention,
         paged_attention,
         reshape_and_cache,
     )
 elif SYSTEM == "rocm":
     from .rocm import (
-        PREFILL_IN_KV_CACHE,
         SUPPORTS_WINDOWING,
         attention,
         paged_attention,
         reshape_and_cache,
     )
 elif SYSTEM == "ipex":
     from .ipex import (
-        PREFILL_IN_KV_CACHE,
         SUPPORTS_WINDOWING,
         attention,
         paged_attention,
@@ -40,7 +37,6 @@
     "attention",
     "paged_attention",
     "reshape_and_cache",
-    "PREFILL_IN_KV_CACHE",
     "SUPPORTS_WINDOWING",
     "KVCache",
     "Seqlen",