feat: add fynd-benchmark generate-requests subcommand

[tvinagre]

Adds a generate-requests subcommand to fynd-benchmark that produces a per-chain synthetic swap-request dataset (JSON) for capacity load testing. The existing download-trades dataset is Ethereum-only; this covers other chains.

What it does

• Fetches token and pool-count data from Tycho via the new fynd_rpc::protocols::fetch_token_pool_stats, which reuses the pool-count query behind derive-connector-tokens (counts pools per token, joins Tycho token metadata for symbols and decimals).
• Samples token pairs weighted by pool count, excluding same-token pairs, using a seeded StdRng so output is deterministic for a given --seed and token set.
• Log-spaces amounts across ~4 orders of magnitude around one whole token, scaled by each token's decimals.
• Writes a JSON array loadable by the existing --requests-file loaders (load_request_templates / load_requests_from_file).

CLI

fynd-benchmark generate-requests --chain base [--tycho-url <url>] \
  --top-n-tokens 50 --num-requests 2000 --seed 42 --output trades.json

Accepted --chain values: ethereum, base, unichain, bsc, arbitrum, polygon (case-insensitive). zksync parses but has no default Tycho URL, so pass --tycho-url. The Tycho URL defaults to the chain's Fynd endpoint; auth uses TYCHO_API_KEY.

Tests and validation

Sampling and serialization are covered with fixture token data (no network): amount scaling, weighted pair sampling with no self-pairs, seed determinism, minimum-token guard, and a round-trip that loads the generated dataset back through load_request_templates.

The Tycho fetch is a thin network layer. Against a hosted endpoint it was exercised only up to the authentication boundary (the RPC call is made and auth is enforced). Live end-to-end generation against ethereum and base is validated in-cluster, where a one-off Job in the staging-fynd namespace runs the command with TYCHO_API_KEY mounted from the existing secret, so the key never leaves the cluster.

🤖 Generated with Claude Code

[github-actions]

No API Breaking Changes Detected

The PR title signals breaking changes, but cargo-semver-checks found none.
If the breaking change is behavioral, CLI, or config-level (not public Rust API), this is expected.
Otherwise, consider using fix: instead of feat: in the PR title.

[tvinagre]

Two robustness fixes from production use on 2026-07-15 (commit e782f59):

Client-side fetch timeout. Against a degraded indexer (tycho-base-beta), the per-protocol-system component fetches hung indefinitely (18+ min byte-flat stalls). Each Tycho component fetch and the token-metadata fetch now run under a client-side deadline. New --fetch-timeout-secs flag (default 300s per fetch). On timeout the run fails fast with an error naming the protocol system that stalled and suggesting --protocols to exclude it, rather than hanging.

--min-tvl flag. The dedicated tycho-fynd-* indexers plan-gate component queries ("tvl_gt parameter is required on this plan"), so generate-requests previously could not query them at all. New optional --min-tvl passes tvl_gt to the component query (reusing the same with_tvl_gt mechanism the scale subcommand's --min-tvl builds on). The flag help notes that tycho-fynd-* endpoints require it.

Both new knobs are carried on a new TokenPoolStatsParams struct threaded into fynd_rpc::protocols::fetch_token_pool_stats.

Verification (nightly-2026-06-28): cargo fmt --check, clippy --all --all-features --all-targets -D warnings, and rustdoc all clean; cargo nextest run -p fynd-rpc -p fynd-benchmark = 126 passed. New tests cover the timeout error path (deadline reported, inner error surfaced, success passthrough) and CLI parsing of both flags.