2026-09-01 (/loop, user task) - orbistoun-fs test isolation + clippy debt fixed

Root cause of the intermittent sendfile failure and the socket hang was the same landmine: descriptor::write special-cased fd == 4 as the kernel R/W escape pipe (swallowing the write, returning success) before consulting the table - so once an escape test set the escape address, any later test whose real file/socket descriptor was 4 had its writes silently discarded. Fixed by checking the table first: a real descriptor at 4 is written; only an un-tabled 4 (the actual pipe) is swallowed.

Also made exclusively() reset the process-wide statics it only serialised before (descriptor, fcntl, directory, mount tables + the escape address), and added the exclusively() guard to the socket and fcntl tests that touch shared state without it (plus a bounded connect/read timeout on the socket test so it can never hang). 14+ consecutive clean full-suite runs.

Cleared the escape.rs/socket.rs clippy debt in the same pass: long-literal separators, a split of the two-unsafe-op setsockopt block with real SAFETY comments, and a refactor of read_kernel_pipe (135-> <100 lines) by extracting dynlib_obj (the three near-identical dynlib_obj blocks) and a copy_from helper (the per-region copy tail). Also fixed the clippy regression my write-match change introduced. orbistoun-fs clippy is clean; escape tests still pass (behaviour preserved).