perf: use a reusable bytes.Buffer to speed up MutableTree.String

[odeke-em]

The prior code used a naive string concatenation
str += "..."

which is very slow and inefficient especially when iterating over
all the keys, but also it performed an unnecessary byteslice->string
conversion for all arguments inside (nodeDB).traverse* using

str += fmt.Sprintf("%s: %x\n", string(key), value)

notice the string(key) passed into fmt.Sprintf("%s"?
At bare minimum that code should just simply be:

str += fmt.Sprintf("%s: %x\n", key, value)

per https://twitter.com/orijtech/status/1462100381803102213?s=20
which saves a whole lot of cycles and RAM.

This change uses:

• (*bytes.Buffer) in combination with fmt.Fprintf
• A sync.Pool with (*bytes.Buffer) values to reuse buffers

and the results are profound:

$ benchstat before.txt after.txt
name          old time/op    new time/op    delta
TreeString-8     111ms ± 8%       4ms ± 4%  -96.01%  (p=0.000 n=20+20)

name          old alloc/op   new alloc/op   delta
TreeString-8     734MB ± 0%       2MB ± 1%  -99.72%  (p=0.000 n=20+20)

name          old allocs/op  new allocs/op  delta
TreeString-8     37.0k ± 0%     28.7k ± 0%  -22.40%  (p=0.000 n=20+20)

Fixes #455

[tac0turtle]

lgtm,

[ValarDragon]

Nice catch!

I'm a bit skeptical of the sync pool being helpful here since I don't get why were going to multiple calls to string.

However, this change is clearly much more helpful than not having it! And it'll be pretty easy to see if theres any overheads due to sync pool in the future

[tac0turtle]

@odeke-em can you update this branch with master then I can merge it

[odeke-em]

@odeke-em can you update this branch with master then I can merge it

Done, thank you @marbar3778!

Thank you @ValarDragon @marbar3778 for the reviews!

[robert-zaremba]

Probably we could optimize it even further with strings.Join.