refactor: co-locate 17 more std module tests in std/<module>/

[paul-hammant]

Continues #1634. Seventeen more specs move out of tests/regression/ to
sit beside the modules they cover: audio, bytes, cbor, clapae, config,
encoding, hash, language, message, msgpack, number, regex, schema,
strbuilder, time, url, worker.

All 17 are pure R100 renames — no test content changed.

Selected by import, not by filename

A matching name is not evidence a test belongs to a module, so each
candidate was checked for actually importing its namesake. All 17 do.

test_message looked like a counter-example — it appeared to import only
map and string — until I noticed it imports via backticks
(import std.`message`(*)), which my first grep missed.

Deliberately not moved

About 45 regression tests import exactly one std module but are
compiler/codegen tests that merely use it: test_string_leak_*,
test_return_escape_*, test_closure_captures_ptr_and_string, and
similar. They exercise the heap tracker and closure lowering, not
std.string's API — filing them under std/string/ would misattribute
them. That judgement call is left for a human rather than guessed at.

The reference sweep — the thing I got wrong last time

In #1634 I grepped the unrenamed basenames and missed
emit_c_determinism's two pinned corpus entries; CI caught it. This time
the sweep was exhaustive: every old path across the whole tracked tree,
plus a targeted check of every file that pins a fixed file list.

Six references turned up. Four are comments/docs that were repointed so
they stay accurate — tests/regression/test_result_migration.ae,
tests/runtime/test_runtime_strbuilder.c,
tests/regression/test_std_strbuilder_v2.ae, std/schema/README.md.
None were build wiring, so nothing would have broken, but a comment
naming the wrong path is a trap for the next reader.

CHANGELOG.md and CHANGELOG-archive.md also name old paths and are
deliberately left alone: they record where those files were at the
time.

Verified

• All 17 pass from their new homes — and passed at baseline before the
  move, so the move is the only variable
• make test-ae discovers all 17, and the total holds at 965 — the
  risk with a move like this is tests going quiet rather than failing
• Remaining 7 failures are the known environmental set (h2,
  background-server, Windows-cross, wycheproof); none name a moved test
  (an apparent url hit is curl in the h2 stress test — std_url_test_url
  itself PASSes)
• Staged install ships 0 test files with all 136 modules present,
  so refactor: co-locate std module tests in std/<module>/ #1634's std/ trim covers these automatically with no Makefile change

🤖 Generated with Claude Code