feat: add local_compilation mode for explorer-unverified contracts

[arwer13]

Contracts without explorer-verified source can now be checked at the bytecode level: local_compilation.inputs maps an address to a local solc standard-JSON input (source manifest + compiler settings), and local_compilation.compiler pins the solc build. Source contents are still fetched from the pinned GitHub commit, and the source diff is skipped for these addresses since there is nothing explorer-side to diff against.

Relation to feat/local-compilation-mode

This PR implements the same feature as in feat/local-compilation-mode and keeps its config
schema verbatim (local_compilation: {compiler, inputs}), so any config written
for either branch works with both. We propose landing this implementation because:

1. Rebased on current main. The earlier branch forked ~82 commits back,
   before the allowed_diffs machinery and related process_config refactors,
   and no longer applies cleanly. This one is written against today's main and
   composes with allowlists, caching, and the summary/suggestion output.

2. Local contracts reuse the explorer normalization path. The payload is built
   via the existing _build_contract_payload/_attach_contract_metadata helpers
   instead of by hand, so EVM-version normalization and library re-keying to the
   definition file (required for correct solc linking) apply to local inputs
   exactly as they do to explorer bundles.

3. More guardrails. All-local configs need no explorer hostname/chain-id/token
   at all; unknown addresses in local_compilation.inputs are flagged; a config
   whose local contracts would get zero verification (bytecode comparison off)
   warns; the skipped source diff is logged per contract rather than silent;
   malformed/empty solc inputs are rejected with clear errors.

4. Docs and tests included. README section (schema + why a full-bytecode match
   incl. the CBOR metadata tail is sound evidence for guessed settings), CLAUDE.md
   schema note, and tests covering error paths (malformed JSON, empty sources,
   absolute/relative input paths).

[tamtamchik]

The issue is that if you have sources to create an identical bytecode, you can put it to etherscan / blockscout / sourcify as verified. So I don't think this PR is actually worth to be included.

[TheDZhon]

Closing as the use case seems to be too exotic.