Remove unnecessary dependencies brought in by substrate upgrade

[teor2345]

PR #3394 added significant number of new dependencies, which appear to be duplicate or unnecessary.

We need to:

1. check if each dependency is actually used (cargo tree or cargo tree --all-features)
2. see if we can disable it by patching our fork, upgrading our direct dependencies, or disabling features
3. submit those patches to upstream if needed

Possibly unnecessary dependencies:

• cumulus*
  • only used by runtime-benchmarks feature via frame-benchmarking-cli
• kvdb-rocksdb, which I think brings in: bindgen, cexpr, clang-sys, cxx, cxx-build, librocksdb-sys, rocksdb, bzip2, libz-sys, ruzstd

From PR #3404:

• icu_normalizer, which brings in: icu_collections, icu_provider_macros, icu_properties* , icu_locid*, litemap, tinystr, utf8_iter, yoke, yoke-derive, zerofrom*, zerovec*
  • specifically, utf16_iter, write16, and most of the ICU properties (if they have features) shouldn't be needed for UTF-8 DNS handling
  • unfortunately, there are no features to disable these dependencies in libp2p or hickory-resolver, so trying to add them is not worth the effort

Duplicate dependencies:

• frame-metadata (x3)
  • 15.1.0 is only used by the runtime-benchmarks feature via frame-benchmarking-cli and scale-value
  • 16.0.0 is only used by the runtime-benchmarks feature via frame-benchmarking-cli and subxt*
• itertools (x3 -> x4)
  • 0.10.5 is used by frame-support 39.0.0, criterion 0.5.1, etc
  • 0.11.0 is used by frame-support 39.0.0, sc-network 0.48.0 etc, and benchmarks
  • 0.12.1 is used by sc-network 0.48.0, ouroboros 0.18.4, etc
  • 0.13.0 is used by sc-network 0.48.0, etc
• jsonrpsee* (x3)
  • jsonrpsee v0.22.5 is only used in benchmarks
  • jsonrpsee v0.23.2 is not actually used
  • jsonrpsee v0.24.8 is used consistently by our production RPC code
• keccak vs keccak-hash
  • keccak is used consistently by our production code
  • keccak-hash is only used in benchmarks
• libp2p*
• primitive-types
  • primitive-types v0.12.2 is only used in benchmarks
• prost*
  • upstream duplication in litep2p 0.9.0
• secp256k1*
  • 0.27 is not actually used
• security-framework*, which probably brings in: core-foundation
  • security-framework v2.11.1 is used by litep2p 0.9.0, async-nats 0.37, and jsonrpsee (multiple versions)
• smol
  • smol v1.3.0 is used by libp2p (multiple versions)
• soketto
  • soketto v0.7.1 is only used in benchmarking
• sp-crypto-hashing
• subtle
  • subtle v1.0.0 is used by sc-cli, etc
  • subtle v2.6.1 is used by sc-client-api, etc
• tokio-rustls x4
  • tokio-rustls v0.23.4 is used by actix-web
  • tokio-rustls v0.24.1 is used by litep2p 0.9.0
  • tokio-rustls v0.25.0 is only used in benchmarks
  • tokio-rustls v0.26.0 us used by async-nats, reqwest, sc-offchain, etc
• unicode-width
  • unicode-width v0.1.14 is used by substrate-wasm-builder, sc-informant, sc-tracing, etc
  • unicode-width v0.2.0 is only used in benchmarks
• windows-sys: which brings in: windows-targets (x4, existing duplicates, probably one of the causes of slow Windows builds)
  • windows-sys v0.45.0 is used by sc-executor, wasmtime, etc
  • windows-sys v0.48.0 is used by smol, sc-executor, wasmtime, etc
  • windows-sys v0.52.0 is used by sc-informant, sc-service, substrate-wasm-builder, etc
  • windows-sys v0.59.0 is used by libp2p, smol, hwlocality, jsonrpsee, etc

Note: cargo metadata and tools that use it are unreliable, because of rust-lang/cargo#10801

[nazar-pc]

libp2p duplicate can finally be resolved by switching out patched version to libp2p/rust-libp2p#5692 which minimizes API breakage comparing to the version I had prior to this. At least temporarily, Substrate is again behind with libp2p 0.55.0 release a few weeks ago and who knows when it will be upgraded there.