build: cache the go build and module caches

Nothing was caching them, so every run compiled the whole generated plc4go tree
from cold. On the macos runner that took 105 minutes against 23 for linux and 22
for windows - and none of it was the tests, which come in at about 210 seconds
on all three. The time went on compilation: half an hour for mockery to parse
the packages, half an hour compiling protocols/cbus/readwrite/model for a test
that then ran in 510ms, and so on.

The Go SDK comes from the maven wrapper plugin rather than setup-go, so nothing
brings a cache of its own along. Pinning GOCACHE and GOMODCACHE into the
workspace lets one cache step serve all three runners, whose default locations
would otherwise differ, and picks up the toolchain download with them.

.github/workflows/go-platform.yml

diff --git a/.github/workflows/go-platform.yml b/.github/workflows/go-platform.yml
index 1b39604..21469b2 100644
--- a/.github/workflows/go-platform.yml
+++ b/.github/workflows/go-platform.yml
@@ -79,6 +79,11 @@
   PLC4X_TEST_TRACE_EXECUTOR_WORKERS: ${{ github.event.inputs.traceExecutorWorkers }}
   PLC4X_TEST_TEST_TRANSPORT_INSTANCE: ${{ github.event.inputs.traceTestTransportInstance }}
   DEVELOCITY_ACCESS_KEY: ${{ secrets.DEVELOCITY_ACCESS_KEY }}
+  # Pinned into the workspace so one cache step covers all three runners: the default
+  # locations differ per OS, and the Go SDK here comes from the maven wrapper plugin
+  # rather than setup-go, so there is nothing else caching them.
+  GOCACHE: ${{ github.workspace }}/.gocache
+  GOMODCACHE: ${{ github.workspace }}/.gomodcache
 
 jobs:
   test:
@@ -119,6 +124,25 @@
           restore-keys: |
             ${{ runner.os }}-maven-
 
+      - name: Cache Go build and module cache
+        uses: actions/cache@v6
+        with:
+          path: |
+            ${{ github.workspace }}/.gocache
+            ${{ github.workspace }}/.gomodcache
+          # Without this every run compiles the whole generated tree from cold, which the
+          # macos runner takes about 100 minutes over against 23 for linux.
+          #
+          # Keyed on go.sum alone rather than per commit: these caches run to a couple of
+          # gigabytes each, and the repository only gets 10 for everything including the
+          # maven caches above, so an entry per commit would evict itself immediately. The
+          # cost is that the entry is written once per go.sum and not refreshed after, which
+          # suits this tree - the expensive part to compile is the generated code, and that
+          # changes far less often than the code that uses it.
+          key: ${{ runner.os }}-go-${{ hashFiles('plc4go/go.sum') }}
+          restore-keys: |
+            ${{ runner.os }}-go-
+
       - name: Set Platform options
         id: platform_opts
         uses: actions/github-script@v9.0.0