Files named for an audience, renamed for their content

Two handover documents were named for who they were written to, which is what made them age: a handover is addressed to a moment, and reads as stale the day after it is read. Neither one's content was perishable, though - only the framing.

• HANDOVER-ORBISTOUN.md → docs/LOADING.md, "Loading an obSCEne module: what a loader has to handle". Every observation in it came from several loaders and applies to any of them; nothing in it was ever specific to the one that prompted writing it down.
• HANDOVER-ORBISTOUN-NET.md → docs/CLIENT.md, "Writing a client for the obSCEne protocol". PROTOCOL.md is the specification - what the wire carries; this is the other half - what a client author has to do about it. Consumer-specific references removed: the file now names no project but this one.

Kept as separate files rather than folded into PROTOCOL.md and MODULE-FORMAT.md. Spec and guidance are different documents with different readers, and merging 365 lines of client guidance into a 441-line specification produces something nobody reads end to end.

docs/README.md listed five documents out of seventeen, which is the more damaging version of the same problem. An index reads as complete, so a reader who does not find a subject in it concludes the project has nothing on the subject - two thirds of the documentation was invisible to anyone starting where the index sent them.

Rewritten in full, and gated: doccheck.py now fails if a document exists without a pointer, or if a pointer names a file that does not exist. The second direction is the one that catches a rename leaving a dead link in the first document anybody opens - exactly what today's two renames would have done. Shown to reject in both directions before being believed.