Update routes pre computation strategy to pre compute less routes

[OlivierHecart]

Routes pre computation strategy was a bit brutal and was computing too many unnecessary routes leading to high CPU consumption during system setup.
The routes pre computation strategy was updated to only compute

• data routes for known keys that have no wildcard and have subscribers
• query routes for known keys that have no wildcard and have queriables

In a 30 talkers/30 listeners zenoh_rmw test the stabilisation time was reduced from 35 seconds to 7 seconds.

The p2p throughput and latency tests have been run on a MacBook Air M2. No observed regressions.

[Mallets]

expr() returns &str, correct?
If expr() would return keyexpr you could use

zenoh/commons/zenoh-keyexpr/src/key_expr/borrowed.rs

Line 133 in f90587b

pub fn is_wild(&self) -> bool {

and rely on the key-expr logic instead of using contains() everywhere. It's more a matter of concern encapsulation than anything else.

[OlivierHecart]

I fully agree. The problem is that:

• the resource tree cannot store keyexprs for now without a big refactor.
• doing a try_into keyexpr each time accessing the expr when not needed has a significant cost.

So until we refactor the resource tree, it's far more efficient to do it like this.