Add a context::clock WASM intrinsic

[maoueh]

What

Adds a context WASM host module with a single intrinsic, context::clock, which writes the current block clock into guest memory as an encoded sf.substreams.v1.Clock:

#[link(wasm_import_module = "context")]
extern "C" {
    fn clock(output_ptr: *mut u8);
}

It follows the same calling convention as the state getters — the host allocates the payload on the guest heap and writes a {ptr, len} pair at output_ptr. It returns nothing: unlike state::get_at there is no "not found" case, the clock always exists. A marshalling failure is raised as non-deterministic, so a transient failure is never cached into the store.

Implemented for wasmtime (the default runtime) and wazero (which also covers wasip1/tinygo-v1). The JavaScript/v8 runtime is not covered.

context also joins env, state and logger as a namespace WASM extensions cannot register into, keeping it available for whatever belongs in it next.

Why

Until now the only way for a module to reach the clock was declaring source: sf.substreams.v1.Clock as a module input, which forces the dependency into the module graph even when the clock is needed only occasionally. An intrinsic is callable at any point during execution and costs nothing when unused.

Scope note: network id was dropped

The branch name says network-id because it started as "expose the deployment's network id to modules", with the clock added alongside. The network id half is not in this PR. No compelling use case turned up for it, and a networks: block in the manifest already parameterizes a package per network, which covers the same ground.

The investigation is worth recording, since it also shows the network id would have been awkward to deliver:

• firehose-core already resolves the right value — --advertise-chain-name is looked up in the networks registry, canonicalized from aliases, and cross-checked against the genesis block id. It just never reaches the substreams tier1 app.
• That flag is optional. Without it the chain is inferred from the genesis block during InfoServer.Init, which needs the block stores to be reachable and so resolves well after tier1 config is built.
• Most large deployments rely on that inference: in sf-operator, 21 of 46 environments set advertise_chain_name, and the ones that don't include eth-mainnet, eth-hoodi, eth-sepolia, arb-one, eth-bsc-mainnet and eth-polygon-mainnet.

So any tier1-side network id would have started out unset on the busiest deployments, ruling out failing fast on a missing value and leaving a degraded mode to design around. That's the constraint to solve first if the feature comes back.

Testing

• Unit test: Call.DoClock round-trips through protobuf.
• End to end: a new assert_context_intrinsics_0 module in the complex_substreams testdata package asserts, inside the WASM, that the intrinsic matches the Clock it received as a module input. A mismatch panics in the module, so a true output for every block is the proof. Exercised over blocks 0-10 via the tier2 integration test, on both wasmtime and wazero.

The committed .spkg is rebuilt, which shifts every module hash in that package.

Notes for reviewers

• The first commit is a prerequisite, not part of the feature: the complex_substreams testdata crate could not compile at all, because its generated pb/mod.rs declared four modules whose .rs files are absent from the directory. That had to be fixed before the package could be rebuilt.
• Ergonomic Rust bindings (substreams::context::clock()) belong in substreams-rs and will follow separately. Until then modules declare the extern by hand, as the test fixture does.
• Purely additive: no config, no proto change, no behaviour change for existing modules.

[UlysseCorbeil]

Solid, lgtm