jsonrpc: panics in RPC handlers/validators crash the connection or vanish silently in batch requests

[Lepsina]

Description

jsonrpc.Server.handleRequest does not recover from panics raised by RPC method handlers or by parameter validators (s.validator.Struct(...), called from buildArguments/validateParam). There's already a TODO in the code:

// TODO: add recover() to catch panics from handlers/validators and return a JSON-RPC internal error
// instead of crashing the HTTP connection

There is a problem with both a single request and a batch request.

1. Single (non-batch) request: handleRequest runs directly in the HTTP/WebSocket connection's goroutine. An unrecovered panic propagates up to net/http's own recover and terminates that one connection. With logs but without valid JSON-RPC response is returned to the client.
2. Batch request: each element is processed via s.pool.Go(...). handleBatchRequest never calls s.pool.Wait(), so a panic is silently swallowed: the panicking element's response is simply missing from the batch array, with no error and nothing in the logs

Acceptance criteria

• A panic raised inside an RPC method create JSON-RPC Internal error response instead of crashing the connection.
• A panic raised during validateParams is recovered the same way.
• For notification requests per the JSON-RPC 2.0 spec it should still be logged, just not answered.
• A panic in one element of a batch response must show up as an Internal error for that element, while the rest of the batch completes normally.

Documentation link(s)

• about notifications in JSON-RPC 2.0: https://www.jsonrpc.org/specification#notification
• about panic propagation in pool: https://pkg.go.dev/github.com/sourcegraph/conc/pool
• about panic in http: https://pkg.go.dev/net/http#pkg-overview:~:text=Request%2E-,If,ErrAbortHandler%2E