Per-request timeout override and cancellation token dependency in Insomnia workspaces

Insomnia exposes a global request timeout adjustable per workspace between 1 and 600 seconds, but the logic governing per-request timeout overrides when the underlying request pipeline does not expose cancellation tokens is not formally documented. This ambiguity affects whether the per-request value takes precedence, is silently clamped, or triggers fallback behavior, particularly for requests routed through proxies or targeting long-running endpoints. Additionally, the client-side connection termination strategy when a timeout is enforced—whether via request tab closure, abort endpoint invocation, or operating system signal—is not uniformly documented across the desktop and web client variants, raising the possibility of orphaned socket resources if the OS does not promptly release them.

How does Insomnia resolve per-request timeout precedence without cancellation token support in the request pipeline? What distinguishes client-side connection teardown from OS-level cleanup across desktop and web variants when a timeout expires? Does the timeout error code and retryable flag consistently reflect server response semantics across both client versions?