fix(stream): key stream metadata by namespace-qualified name (#46)

[oreofeolurin]

Fixes #46. Found while verifying #42 items 2/3.

The bug

Stream metadata — partition count plus all three retention settings — was keyed by the bare stream name, while stream_names was already keyed by the namespace-qualified one. Two namespaces holding a same-named stream shared a single metadata entry, so creating the second silently rewrote the first's configuration:

$ flo stream create solo --partitions 4 -n ns1
$ flo stream info solo -n ns1
  Partitions: 4

$ flo stream create solo --partitions 1 -n ns2      # unrelated namespace

$ flo stream info solo -n ns1
  Partitions: 1        ← silently changed

Partition count drives routing, so a tenant could re-route another tenant's stream just by picking the same name — no error, no log line. Retention collided the same way, letting one namespace shorten another's and delete its data.

The fix

Metadata is qualified at every read and write, matching what the name registry already does.

Two distinct read paths needed it, not one. Qualifying the key fixed stream info, but stream list still reported the wrong count — it strips the namespace prefix before looking metadata back up, and it's served by the ShardWalker (serializeWalkStreamNames), not by handleList. Both serializers now take the namespace and re-qualify, the same shape queue_list, processing_list and action_list already use. I'd have shipped a half-fix if I'd only checked info.

deleteStream removes the qualified key and still sweeps the raw one, so entries written by earlier builds don't leak.

Verified

$ flo stream create solo --partitions 4 -n ns1
$ flo stream create solo --partitions 1 -n ns2

$ flo stream info solo -n ns1     → Partitions: 4
$ flo stream ls          -n ns1     → solo  4
$ flo stream info solo -n ns2     → Partitions: 1
$ flo stream ls          -n ns2     → solo  1

$ flo stream alter solo --retention 24 -n ns1
$ flo stream info solo -n ns1     → Retention: Age: 86400s
$ flo stream info solo -n ns2     → (no retention)

Tests

Three e2e tests: partition count via info, partition count via list (the separate path), and retention isolation. All three fail against the unfixed code.

• test-unit ✅ · test-integration ✅
• test-e2e 576/580 — the 4 failures are the kv/cluster set, confirmed failing identically on a stashed baseline.

Note on stacking

Branched off dev, independent of #45 and #47. It touches src/stream/handler.zig near #45's handleInfo change, so expect a small conflict depending on merge order — happy to rebase whichever lands second.