Rename protocol fees to program fees, and "locks" test names to what they check

[mikemaccana]

Follows quicknode/solana-book#190, which removes "protocol" and vague "lock" wording from the book. The book can only name identifiers that exist here, so the rename lands here first.

Identifiers

Before	After
Pool.protocol_fees (perpetual futures, all variants)	program_fees
Reserve.accumulated_protocol_fees (lending)	accumulated_program_fees
collect_protocol_fees / CollectProtocolFees / handle_collect_protocol_fees (lending)	collect_program_fees / CollectProgramFees / handle_collect_program_fees
protocol_fees_accrue_and_owner_can_collect (lending tests)	program_fees_accrue_and_owner_can_collect
proof_rounding_is_protocol_favourable (lending Kani)	proof_rounding_is_program_favourable
place_bid_locks_quote_in_vault, place_ask_locks_base_in_vault (order book)	place_bid_moves_quote_into_vault, place_ask_moves_base_into_vault
test_write_call_locks_the_underlying (options)	test_write_call_moves_underlying_into_vault

Renaming the lending instruction changes its Anchor discriminator, and the account fields get new IDL names. The byte layouts are unchanged.

Prose

• Comments and READMEs that called the program "the protocol" now say "the program". This covers its fees, rounding in its favour, and who keeps the dust.
• "Source of truth" comments now say which field the program reads.
• Left alone:
  • Comments about the Solana protocol lowering slot times.
  • CHANGELOG history entries.

Checks

• cargo fmt --check is clean in every touched crate.
• cargo check --tests passes in every touched anchor, anchor-v1, quasar and kani-proofs crate. The Anchor ones were checked against placeholder .so files because cargo-build-sbf isn't available locally, so CI runs the tests for real.

🤖 Generated with Claude Code

https://claude.ai/code/session_01JGEoAUjMm7Evv69k46eNcn

---

Generated by Claude Code

[mikemaccana]

The anchor-v1-group-1 failure is deepest_path_adds_little_compute_to_insert_fill_and_cancel in finance/order-book/anchor-v1. An insert at depth 64 costs 15,228 CU more than on a shallow book, and the test allows 15,000.

I don't think this PR causes it:

• The only change this PR makes to the order-book program is a comment in place_order.rs, so the program it builds is the same as main's. The other changes rename tests.
• The Anchor v2 copy of the same test passed on this PR.
• The test landed on 23 September (fb46fde) and passed in the 28 September scheduled run.
• Cargo.lock is gitignored (**/*/Cargo.lock), so each CI run resolves the dependencies afresh. A newer patch release of a dependency since 28 September would change the compute cost without any change to this repository. That's my best guess, but I haven't confirmed it: I couldn't install the Solana toolchain to reproduce it.

No fix exists yet. I'm re-running the failed job once. If it fails again, the fix is to find which dependency moved and either pin it or re-measure the limit, in its own PR. I won't raise the limit here to get this PR green.

---

Generated by Claude Code