Make derived cells lazy

[fewkz]

right now, deriving a formula from a cell will result in that calculation being made whenever the cell changes, even if nothing is subscribed to that formula. We should make it so formulas are lightweight and lazy, which is the behavior in the signals library cells is based off of: https://github.com/preactjs/signals

[fewkz]

So for example, for the follow code:

local c = cell(0)

local f1 = formula(function()
    print(`f1 updated`)
    return `f1: {c.value}`
end)

local f2 = formula(function()
    print(`f2 updated`)
    return `f2: {c.value}`
end)

local isone = formula(function()
    return true
end)

subscribe(function()
    if isone.value then
        print(f1.value)
    else
        print(f2.value)
    end
end)

We want the following behavior:

-- prints f1 updated
-- prints f1: 0
c.value = 10
-- prints f1 updated
-- prints f1: 10
c.value = 20
-- prints f1 updated
-- prints f1: 20
isone.value = false
-- prints f2 updated
-- prints f2: 20

However, currently the behavior is as follows

-- prints f1 updated
-- prints f2 updated
-- prints f1: 0
c.value = 10
-- prints f1 updated
-- prints f2 updated
-- prints f1: 10
c.value = 20
-- prints f1 updated
-- prints f2 updated
-- prints f1: 20
isone.value = false
-- prints f1 updated
-- prints f2 updated
-- prints f2: 20

Clearly, formulas are calculated despite not being needed. As well, this means that merely CREATING a formula has the overhead of performing its calculation. Having it be deferred to its initial subscription would be better.

[fewkz]

This isn't necessarily a critical issue, as cells still works perfectly fine without it, and I don't particularly feel like working on adding this right now, but implementing this should be a simple addition that will just optimize any code using cells without requiring changes.

[fewkz]

You could also accomplish this behavior by just not using formula() and just creating a function that operates on the cell's value directly. This could technically be faster for some calculations since there's no overhead from formula. You could consider it like inlining the formula. However, I'm sure there's still cases where you would want to have the result of a calculation cached in the dependency graph.