fix: invoke onRequest callback for streaming prediction requests

[mmattu-wd]

Issue

The onRequest callback prop ((request: RequestInit) => Promise<void>) is not called when sending streaming prediction requests.

This breaks any consumer relying on it to inject auth headers, set credentials: 'include', or mutate the request before it is sent.

Root cause:
The streaming path in fetchResponseFromEventStream calls fetchEventSource from @microsoft/fetch-event-source directly, and never invokes onRequest. All other request paths (config fetch, stream availability check, non-streaming prediction, TTS) go through sendRequest() in src/utils/index.ts, which correctly calls onRequest(requestInfo) before each fetch().

This means GET requests like getChatbotConfig and isStreamAvailableQuery receive injected headers/credentials, but the streaming POST to /api/v1/prediction/{chatflowid} does not — causing auth failures for users who authenticate via cookies or bearer tokens.

Resolution

fetchEventSource accepts a fetch parameter (FetchEventSourceInit.fetch?: typeof fetch) that replaces its underlying fetch implementation. A customFetch wrapper is defined inside fetchResponseFromEventStream that:

1. Calls props.onRequest(init) if the callback is provided — giving the consumer a chance to mutate the RequestInit (add headers, set credentials, etc.)
2. Delegates to native fetch with the (potentially mutated) init object

This mirrors the existing pattern in sendRequest() and generateTTSQuery(), and requires no changes to any other file.

[gemini-code-assist]

Code Review

This pull request introduces a customFetch wrapper in the Bot component to allow for request interception via an onRequest prop. A critical syntax error was identified in the fetchEventSource URL template literal, where extra quotes and a closing brace were accidentally included, which will lead to malformed request URLs.