fix(llm-bridge): use finish_reason to determine function calling (#1149)

This pull request refactors the chat completion handling logic in
`pkg/bridge/ai/chat.go` to improve streaming function call support and
clarify response processing. The main changes include updating the
`chatResponse` interface and its implementations, enhancing the
streaming response handling, and adjusting related test expectations.

**Chat response interface and implementation improvements:**

* The `checkFunctionCall` method of the `chatResponse` interface now
accepts an `EventResponseWriter` and `chatContext`, allowing for more
flexible response handling during streaming and non-streaming calls. All
implementations of this method were updated accordingly.
[[1]](diffhunk://#diff-e71e367c6ce3678d59b1edf614d34eebab075b2fcef4094e8461e42bcbb095c9L21-R22)
[[2]](diffhunk://#diff-e71e367c6ce3678d59b1edf614d34eebab075b2fcef4094e8461e42bcbb095c9L224-R230)
[[3]](diffhunk://#diff-e71e367c6ce3678d59b1edf614d34eebab075b2fcef4094e8461e42bcbb095c9L383-R424)
* The `streamChatResp` struct was refactored to better manage streaming
function call responses, including new fields for tracking buffered
chunks and tool call deltas, and improved logic for detecting and
processing function calls in a stream.
* The `getToolCalls` method in `streamChatResp` now sorts tool calls by
their index for consistency and correctness, using the new
`slices.SortFunc` utility.

**Test updates:**

* Updated test expectations in `pkg/bridge/test/llm_server_test.go` and
`pkg/bridge/test/service_test.go` to match the new, simplified response
content ("How can I assist you today?").
[[1]](diffhunk://#diff-7cf5047c4d1e5bbcb5c69762b88b717c281a7cf56eb83599aa7968d6484469c1L95-R95)
[[2]](diffhunk://#diff-7cf5047c4d1e5bbcb5c69762b88b717c281a7cf56eb83599aa7968d6484469c1L111-R111)
[[3]](diffhunk://#diff-446f1951a3258c7bf56256ed2316bf6961314ed536089924c55e7f8946d1f756L497-R497)

**Other minor improvements:**

* Added the `slices` package import for sorting functionality.
* Minor struct and method signature cleanups, such as explicit field
assignments and additional context tracking in `chatContext`.

These changes collectively improve the reliability and clarity of chat
completion handling, especially for function call scenarios in streaming
responses.

---------

Co-authored-by: C.C <[email protected]>
Co-authored-by: Copilot <[email protected]>