Tune heuristics for feedback vector allocation

Recently we changed feedback vector allocation to be based on the
bytecode size of the function. The threshold at which the feedback
vectors are allocated was set to 12 * bytecode size of the function.
This caused a couple of regressions on IC:duration and some regressions
on other benchmarks. To avoid these regressions this cl reduces the
scale factor to 8 instead of 12.

Bug: chromium:1187733
Change-Id: I0553d368434499cc52a6e786b5de6d6b954e6546
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2778295
Commit-Queue: Mythri Alle <mythria@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#73624}

src/flags/flag-definitions.h

@@ -520,7 +520,7 @@ DEFINE_BOOL(use_ic, true, "use inline caching")
 DEFINE_INT(budget_for_feedback_vector_allocation, 940,
            "The budget in amount of bytecode executed by a function before we "
            "decide to allocate feedback vectors")
-DEFINE_INT(scale_factor_for_feedback_allocation, 12,
+DEFINE_INT(scale_factor_for_feedback_allocation, 8,
            "scale bytecode size for feedback vector allocation.")
 DEFINE_BOOL(feedback_allocation_on_bytecode_size, true,
             "Instead of a fixed budget for lazy feedback vector allocation, "