perf: optimize portfolio runtime delivery

Replace the blocking asset loader with progressive rendering, defer expensive media and Three.js work, and reduce hydration and animation costs. Move deployment from a static Nginx export to the standalone Next.js runtime with optimized dynamic images.

PERFORMANCE.md

@@ -0,0 +1,36 @@
+# Performance baseline and targets
+
+The pre-change development build blocked the page behind a preload manifest of
+76 images (about 6.49 MiB encoded). Under the recorded throttled mobile run the
+loader remained visible for about 73 seconds and LCP was about 36 seconds. The
+old initial client chunks were about 346 KiB gzip, including a Three.js chunk of
+about 172 KiB gzip.
+
+Measured pre-change section heights used for intrinsic containment:
+
+| Viewport | Projects | Experience | Skills | Education | Milestones | Footer |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| 390 px | 746 | 3776 | 3384 | 1010 | 3688 | 472 |
+| 768 px | 727 | 2786 | 1877 | 1005 | 2338 | 495 |
+| 1024 px | 745 | 2304 | 1391 | 568 | 1830 | 569 |
+| 1440 px | 797 | 2081 | 1233 | 475 | 1665 | 613 |
+
+Cold-cache acceptance targets are initial JavaScript at or below 200 KiB gzip,
+initial transfer at or below 500 KiB before scrolling, LCP at or below 2.5 s,
+INP at or below 200 ms, and CLS at or below 0.1. Compare medians from three
+mobile-profile runs; treat differences inside run-to-run variance as neutral.
+
+## Post-change measurements
+
+The production standalone build references 198,652 bytes (194.0 KiB) of
+initial JavaScript after gzip. Its estimated English first-view transfer is
+391,274 bytes (382.1 KiB), including compressed HTML, JavaScript, CSS, the hero
+avatar, flags, and the two locale-relevant fonts. No font is force-preloaded;
+the browser requests only the faces used by the active locale. The deferred
+Three.js chunk is 131,036 bytes gzip and is not part of the initial script set.
+
+Three fresh 390 x 844 production-browser runs measured LCP at 1,048 ms,
+1,296 ms, and 1,432 ms (1,296 ms median), with CLS 0 in every run. These local
+runs validate the build and relative improvement but are not a substitute for
+field data or a repeatable hosted-network/mobile-CPU lab profile. INP should be
+tracked in production once real interaction samples are available.