fix(crispasr): support long VibeVoice transcripts

Pin CrispASR to the VibeVoice duration-scaled generation budget fix so long diarized transcripts are not truncated at 512 tokens. Fetch the exact pinned revision so the dependent upstream pull request commit is available before merge.

Assisted-by: Codex:gpt-5

backend/go/crispasr/Makefile

@@ -58,7 +58,7 @@ sources/CrispASR:
 	cd sources/CrispASR && \
 	git init && \
 	git remote add origin $(CRISPASR_REPO) && \
-	git fetch origin && \
+	git fetch origin $(CRISPASR_VERSION) && \
 	git checkout $(CRISPASR_VERSION) && \
 	git submodule update --init --recursive --depth 1 --single-branch -- \
 		ggml third_party/c2pa-audio