palladium: map fwrite in LogPerfEndpoint to TB_IMPORT (#589)

When we enable XSPerf for XiangShan, logPerfs will be collected to
LogPerfEndpoint. Previous we map all logPerfs to GFIFO, and
LogPerfEndpoint consumes abount 140M gates, more than XiangShan itself.

This change map fwrite in LogPerfEndpoint to TB_IMPORT, reducing gates
from 140M to 68M.

Note we cannot map all fwrites to TB_IMPORT, as some fwrites generated
by chisel assert will include complex signal, greatly increase compile
time from about 4 hour to more than 1 day.

scripts/palladium/argConfigs.qel

@@ -4,3 +4,4 @@
 * $fatal TB_IMPORT
 * $random TB_IMPORT
 * $fwrite GFIFO
+LogPerfEndpoint $fwrite TB_IMPORT