feat: panic recovery in handlers

[mcriley821]

Summary

A panicking handler currently propagates the panic up through handleRequests, crashing the goroutine and tearing down the connection ungracefully. The peer receives no response and no error — the connection simply closes.

Add panic recovery so a panicking handler is converted into an InternalError response (for requests) or silently discarded (for notifications), keeping the connection alive.

Proposed behaviour

• Recover the panic inside the per-request goroutine in handleRequests
• For requests (non-nil ID): reply with NewError(InternalError, ..., nil)
• For notifications (nil ID): discard silently — no response is expected
• Log the recovered panic value via the configured Logger if present

Motivation

• A single misbehaving handler should not take down the entire connection
• Brings parity with github.com/creachadair/jrpc2 and github.com/AdamSLevy/jsonrpc2
• Safer in production where handler panics (e.g. nil pointer) are hard to rule out entirely