fix: better tracking of cache entries

[klkvr]

Currently, during sparse compilation (and probably in some other cases) we might end up in a situation when cache contains empty entries which can actually produce artifacts when compied.

For example, if A.sol imports B.sol and sparse filter tells us to only compile A.sol, then B.sol will appear in cache (because we have to track its source hash to invalidate A.sol's cache), however, we will erase its output selection when invoking solc, so it will not have any artifacts.

Thus, if compiler is invoked later without sparse filter, it will simply skip compiling B.sol because it will assume that it just does not contain any contract definitions:

compilers/src/cache.rs

Lines 658 to 663 in 4cf7843

	// only check artifact's existence if the file generated artifacts.
	// e.g. a solidity file consisting only of import statements (like interfaces that
	// re-export) do not create artifacts
	if entry.artifacts.is_empty() {
	trace!("no artifacts");
	return false;

Because of that, artifact for B.sol will be missing until the cache is cleared.

The issue is basically that we might have two situations:

• Entry does not contain artifacts because it does not produce them when compiled
• Entry does not contain artifacts because we've never tried to compile it

This PR adds seenByCompiler flag which is set to false by default for all newly created entries and switched when entries are compiled as dirty files (without pruned output selection).

[mattsse]

nice, this is indeed ambiguous without tracking this context manually.

this is technically a new cache format,

we could finally bump and rename the ETHERS_FORMAT_VERSION variable (separately)

[mattsse]

we could add a setter fn like with_sparse_output separately

[mattsse]

this is helpful!