ci(nix): retry flake evaluation on GitHub API rate limits

Add a reusable retry helper that detects GitHub API 403 rate-limit errors
during Nix flake evaluation and retries with exponential backoff.

Apply the retry logic to flake-only CI tests so transient GitHub rate
limits no longer cause random CI failures while preserving fast failure
for real errors.

https://chatgpt.com/share/693d7ec5-ac70-800f-a627-ef705c653ba1

scripts/nix/lib/retry_403.sh

@@ -0,0 +1,52 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+if [[ -n "${PKGMGR_NIX_RETRY_403_SH:-}" ]]; then
+  return 0
+fi
+PKGMGR_NIX_RETRY_403_SH=1
+
+# Retry only when we see the GitHub API rate limit 403 error during nix flake evaluation.
+# Retries 7 times with delays: 10, 30, 50, 80, 130, 210, 420 seconds.
+run_with_github_403_retry() {
+  local -a delays=(10 30 50 80 130 210 420)
+  local attempt=0
+  local max_retries="${#delays[@]}"
+
+  while true; do
+    local err tmp
+    tmp="$(mktemp -t nix-err.XXXXXX)"
+    err=0
+
+    # Run the command; capture stderr for inspection while preserving stdout.
+    if "$@" 2>"$tmp"; then
+      rm -f "$tmp"
+      return 0
+    else
+      err=$?
+    fi
+
+    # Only retry on the specific GitHub API rate limit 403 case.
+    if grep -qE 'HTTP error 403' "$tmp" && grep -qiE 'API rate limit exceeded|api\.github\.com' "$tmp"; then
+      if (( attempt >= max_retries )); then
+        cat "$tmp" >&2
+        rm -f "$tmp"
+        return "$err"
+      fi
+
+      local sleep_s="${delays[$attempt]}"
+      attempt=$((attempt + 1))
+
+      echo "[nix-retry] GitHub API rate-limit (403). Retry ${attempt}/${max_retries} in ${sleep_s}s: $*" >&2
+      cat "$tmp" >&2
+      rm -f "$tmp"
+      sleep "$sleep_s"
+      continue
+    fi
+
+    # Not our retry case -> fail fast with original stderr.
+    cat "$tmp" >&2
+    rm -f "$tmp"
+    return "$err"
+  done
+}