chore(api): simplify hub fan-out — drop NATS-style wildcard subscriptions #100

Part of #87. Scoped narrowly: this issue covers only the hub wildcard simplification bullet from #87 — it does not touch raw SQL, pipes, or structured query (those stay as-is for now).

Context

The hub (internal/api/hub.go) supports NATS-style * (one-token) and > (multi-token) wildcards on the subscriber side. After the MVP cuts in #89, every producer publishes a concrete ingest.<table> subject and the SDK always subscribes to one concrete subject. The wildcard fan-out is unused machinery — keep it and we keep paying for it in code, tests, and docs that imply the API is more general than it is.

What to change

internal/api/hub.go

• Delete matchTopic (lines 121-142).
• In Broadcast (lines 62-119): drop the wildcard pattern loop (lines 100-118) and the sent dedup map; keep only the exact-topic delivery loop.
• Remove the now-unused strings import.
• Keep RLock/RUnlock — the lock contract is unchanged.

internal/api/hub_test.go

Delete the eight wildcard tests (lines 212-411):

• TestHub_WildcardGreaterThan
• TestHub_WildcardStar
• TestHub_WildcardStarNoMultiToken
• TestHub_WildcardGreaterThanMultiToken
• TestHub_WildcardDoesNotMatchExact
• TestHub_BareGreaterThanMatchesAll
• TestHub_WildcardNoDuplicateDelivery
• TestMatchTopic

internal/api/stream_sse.go (lines 37-40)

Kill the topic = "ingest.>" default. Decide one of:

• (A) Return 400 when ?topic= is missing.
• (B, recommended) Replace ?topic=ingest.<table> with ?table=<name> and build the subject server-side. This stops leaking the NATS subject convention into the public HTTP API. SDK is a one-line change either way; direct curl users have to switch from ?topic=ingest.clicks to ?table=clicks.

internal/api/stream_ws.go

• ?topic= query-param fallback (lines 125-140): same A/B decision as SSE.
• In-band {\"action\":\"subscribe\",\"topic\":\"…\"} command (lines 149-158): reject any topic containing * or > with an in-band error frame. (Or rename the command field to table if going with B.)

internal/api/stream_test.go

Re-check: any test relying on the wildcard default needs to set an explicit table/topic.

clients/ts/src/stream/ws-manager.ts

• Drop the wildcard dispatch loop (lines 172-180) — already unreachable since client.ts:90 only ever builds ingest.<table>.
• Delete the matchTopicPattern helper (~line 243).
• Update the doc comment at lines 16-24 to remove the "wildcard" hint.
• If option B is chosen: change client.ts:90 to send table instead of topic to match the new server contract.

Docs

• docs/api.md lines 305-306 (table row claiming wildcards) and line 352 — change topic default to "(required)" or rename to table, and remove the "Supports NATS wildcards: * matches one token, > matches one or more remaining tokens" sentence.
• docs/api.md:335 — curl example already uses a concrete topic; just update if renaming.
• docs/architecture.md:93 — "Default topic is ingest.> (all tables)" → describe the new explicit-table contract.
• docs/development.md:96 — ?topic=ingest.clicks example still works; update if renaming.
• CHANGELOG.md under [Unreleased] → Removed (and Changed if renaming): hub wildcard subscriptions; SSE/WS now require an explicit table.

Out of scope (do not touch in this PR)

Per the #87 audit, these are deliberately left alone here:

• /v1/query (raw SQL), wh.sql() in the SDK, policy.RawSQL.
• /v1/pipes/*, /v1/admin/pipes/*, internal/pipes/, WH_PIPES_DIR, wh.pipe() / wh.pipes.*.
• /v1/tables/{table}/query (structured query AST) and internal/query/.

Decisions to lock in before coding

1. Option A or B above? (Recommend B — clean break, no leak of NATS subject names into the HTTP API.)
2. WS in-band command field name — stay on topic (with pattern rejection) or rename to table? Should match the SSE choice for consistency.

Bonus cleanup (optional, dead code already)

clients/ts/src/pipes.ts:40-42 — PipeRef.stream() calls _createStream(pipeName), which subscribes to ingest.<pipeName> for a NATS subject no producer publishes. Dead path regardless of this work; worth removing if you're already in the SDK streaming files.

Estimated impact

• ~150 LOC removed (mostly tests).
• Zero new abstractions, no new tests.
• Breaking change to SSE/WS public API: callers must supply an explicit table/topic (and may need to rename the query param if going with B).
• make ci should remain green — the deleted tests are the only ones exercising wildcard behavior.