feat: support deepseek (#69)

The DeepSeek API reduces costs through disk caching, and DeepSeek-Coder-V2-0724 achieves GPT-4-Turbo-0409 level code capabilities with excellent math and reasoning skills.

lua/avante/config.lua

@@ -6,8 +6,8 @@ local M = {}
 
 ---@class avante.Config
 M.defaults = {
-  ---@alias Provider "openai" | "claude" | "azure"
-  provider = "claude", -- "claude" or "openai" or "azure"
+  ---@alias Provider "openai" | "claude" | "azure" | "deepseek"
+  provider = "claude", -- "claude" or "openai" or "azure" or "deepseek"
   openai = {
     endpoint = "https://api.openai.com",
     model = "gpt-4o",
@@ -27,6 +27,12 @@ M.defaults = {
     temperature = 0,
     max_tokens = 4096,
   },
+  deepseek = {
+    endpoint = "https://api.deepseek.com",
+    model = "deepseek-coder",
+    temperature = 0,
+    max_tokens = 4096,
+  },
   behaviour = {
     auto_apply_diff_after_generation = false, -- Whether to automatically apply diff after LLM response.
   },