docs: day-one client list -- which http-sql clients we supply, and the rule they follow

[ssilvius]

Problem

http-sql has one reference client (TypeScript, ~40 lines) and smugglr's Rust profile. The prior-art research (legion document RESEARCH-HTTPSQL-GRAVEYARD) found that a spec with no popular client loses to a client with no spec: PlanetScale implemented Neon's undocumented endpoint because Neon's driver had users. Operator direction: create the day-one list of clients we supply, so traffic has a way in from every runtime our first segments use.

Acceptance criteria

• A clients.md at the repo root lists the Tier 1 clients that ship before anything is announced, ordered by where http-sql traffic originates, each with runtime, audience, proposed package name, and the reason it is day one.
• The same document states the rule every client follows: fetch and JSON only, the request headers and body shapes, tagged-value handling, error mapping including statementIndex and unknown vendor codes, no sessions or query builder, and the completion criterion (passes the client cases in checks/ against the reference server).
• A per-language table pins what bigint and blob decode to, so no client can round a 64-bit integer and call itself done.
• A Tier 2 list with the reason each is not day one, and a declined list, so absence is not read as an oversight.
• The document says where the clients live and under what license.
• README links to it.

Out of scope

Writing any client. Package names are proposals until the first one is published.