perf(compile): rebalance disk from build variants to ccache

Measured on staging: 46 per-variant build trees held 25 GB - each one a
full copy of the same ESP-IDF component objects - while ccache, the
cache that serves EVERY target and variant, was capped at 8 GB and
permanently full: 99.97% of its limit, 153 evictions, 47% lifetime hit
rate. Objects were being thrown out before they could be reused, so
builds recompiled work they had already done.

Variant cap 12 -> 6 and ccache 8G -> 20G. The asymmetry is deliberate:
an evicted variant dir rebuilds quickly FROM ccache, whereas a ccache
miss recompiles from source. Pruning the cold variants freed 8 GB on
the staging box (88% -> 84% disk) before the larger cache can grow into
it.

Dockerfile.standalone

@@ -308,7 +308,7 @@ ENV CCACHE_NOHASHDIR=1
 # runtime by the volume mount. ccache reads CCACHE_MAXSIZE / CCACHE_COMPRESS
 # / CCACHE_COMPRESSLEVEL on every invocation and they override any conf-file
 # value, so the cap actually applies at runtime.
-ENV CCACHE_MAXSIZE=8G
+ENV CCACHE_MAXSIZE=20G
 ENV CCACHE_COMPRESS=1
 ENV CCACHE_COMPRESSLEVEL=6
 RUN mkdir -p /var/cache/ccache /var/lib/velxio-build /root/Arduino

backend/app/services/espidf_compiler.py

@@ -219,7 +219,13 @@ def _run_with_streaming(
 # coldest. Each variant is a full ESP-IDF build tree (~240 MB). Distinct
 # variants = distinct (board options x resolved library set). The global ccache
 # means an evicted-then-rebuilt variant warms up in seconds.
-_MAX_BUILD_VARIANTS = 12
+# Was 12. Each variant is a FULL ~1 GB build tree holding its own copies of
+# the same ESP-IDF component objects, so 46 of them had grown to 25 GB while
+# ccache - the cache that actually serves EVERY target and variant - sat
+# capped at 8 GB, permanently full (99.97%, 153 evictions, 47% hit rate).
+# Fewer trees, a bigger shared cache: a variant that gets evicted rebuilds
+# quickly FROM ccache, but a ccache miss recompiles from source.
+_MAX_BUILD_VARIANTS = 6
 
 
 def _evict_cold_variants(target_dir: Path, keep: int) -> None: