Upgrade libuv to 1.53.0 via the rescript-lang/luv fork - #8883

The opam override for Luv now fetches commit 8c12b1a of rescript-lang/luv
(PR rescript-lang/luv#2) instead of patching the upstream 0.5.14 tarball.
The fork vendors libuv 1.53.0 instead of 1.48.0 and already carries the
ARM64 musl -mno-outline-atomics fix, so the local patch is gone. The
package is versioned 0.5.14+rescript.1 and dune-project/rescript.opam pin
that version.

Review: Why a git URL and not a release tarball? The fork keeps libuv as a
git submodule; GitHub archives omit submodules, while opam's git fetch
initializes them. The commit hash should move to the merge commit once
rescript-lang/luv#2 lands.

Review: Does the OCaml rewatch use any changed Luv API? Only FS_event,
Loop, Timer, Handle and Error (native_watcher.ml). The fork's API change
(statfs reserved fields read as zero) does not reach it.

Area: rewatch ci
Output: unchanged
Tests: none, because no observable behavior changes; the shared rewatch
 suite (113 checks, incl. watch mode), the rewatch-ocaml focused suite and
 scripts/test.js -ounit -build -mocha -docstrings pass on the new libuv
Measured: watch edit latency, median of 21 interleaved edits (basic
 fixture) 99 -> 100 ms, A/A control 98 ms; 400-directory fixture 98 -> 102
 ms, A/A control 100 ms; watch cold start, median of 7 (400 dirs) 213 ->
 211 ms, A/A 212 ms; idle watcher CPU over 20 s 70-90 ms for all; clean
 build (400 modules), median of 9, 180 -> 180 ms; linux-arm64 devcontainer
 under load, job tmp watch_ab.py and clean_ab.py
Signed-off-by: Christoph Knittel <ck@cca.io>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q9Q6perkzKBQQebDnVXAEK

The devcontainer's postCreate.sh and the CONTRIBUTING setup steps now add
.github/opam-repository as the rescript-overrides repository before
`opam install .`, as the CI setup action does. luv.0.5.14+rescript.1 exists
only there, so without it the solver finds no Luv; this used to go unnoticed
because luv.0.5.14 is also in the default repository.

Review: Is it safe when postCreate runs again? Adding an already registered
name and URL succeeds and changes nothing (checked with opam 2.5.0).

Area: ci docs
Output: unchanged
Tests: none, because the test-devcontainer CI job exercises postCreate.sh
Signed-off-by: Christoph Knittel <ck@cca.io>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q9Q6perkzKBQQebDnVXAEK

rescript.opam (from rescript.opam.template) now pins luv.0.5.14+rescript.1
to the rescript-lang/luv commit via pin-depends, so `opam install .` fetches
it in CI, the devcontainer and local setups alike. The
.github/opam-repository overlay, its registration in the CI setup action,
in postCreate.sh and in CONTRIBUTING, and its cache-key entry are removed;
the cache key still changes with the pin through rescript.opam.

Review: Does --deps-only apply pin-depends? Yes: with no overlay
registered, `opam install . --deps-only --with-test` (opam 2.5.0) pinned
and installed luv.0.5.14+rescript.1 at 2094c025, with libuv 1.53.0.

Area: ci docs
Output: unchanged
Tests: none, because CI's opam setup and the test-devcontainer job
 exercise the pin
Signed-off-by: Christoph Knittel <ck@cca.io>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q9Q6perkzKBQQebDnVXAEK