refactor(core): add an empty table in constants then we can reuse it

[chronolaw]

Summary

The read-only empty table is very useful, it is widely used in kong's code base,
but its definition scatters in many different files.

This PR defines one and only empty table in constants.lua, other lua modules can use it easily.

I just changed EMPTY in some but not all modules,
if most of us think it is good, we will try to change all of them.

Checklist

• The Pull Request has tests
• There's an entry in the CHANGELOG
• There is a user-facing docs PR against https://github.com/Kong/docs.konghq.com - PUT DOCS PR HERE

Full changelog

• [Implement ...]

Issue reference

Fix #[issue number]

[bungle]

Is the EMPTY most of the time used in (to avoid if statements or something):

local x = (d or EMPTY).x or "default"
local x = (d and d.doge or EMPTY).x or "default"

Which you also see in forms of:

local x = (d or {}).x or "default"
local x = (d and d.doge or {}).x or "default"

I am not huge fan of either, but some prefer it. The constant seems to be just to avoid table creation (but I am not sure if in those above examples LuaJIT would optimize it away anyway, and even if not, is it a big deal). The later example has benefit of not requiring a module to get EMPTY and is also shorter, and doesn't leak.

If I were to pick, I would just change code everywhere to just or {}. Without that EMPTY thing in constants or module level global.

[hanshuebner]

I don't see the benefit of an EMPTY constant. It is longer than {} when used and requires an additional declaration. It also reminds me of #define BEGIN {. Let's not do this.

[chronolaw]

We already defined many EMPTY in modules, could this PR reduce the duplicated code?

[bungle]

We already defined many EMPTY in modules, could this PR reduce the duplicated code?

I guess the suggestion was to remove all EMPTYs everywhere and just or {}.

[chronolaw]

Let's keep the code base no changing until we get a agreement.