perf(hashkey): refine the memory reserving for SerializedKey

[st1page]

I think it is not good enough, we can get each row's value encoding size like #8924. And after that, we can avoid any grow of the vec

Originally posted by @st1page in #8885 (comment)

[kwannoel]

More info from @st1page offline:

In nexmark q17, it works well because all the string have the same length and they comes from date, but we need to optimize it in more case :h

[kwannoel]

Additionally, fixed size key should not have estimation, it does not use value encoding. Currently implementation couples both serialized and fixed size key.