refactor(aggregator): accelerate build times

[jpraynaud]

This PR includes the change that stops the aggregator's warp router from being one giant generic type:

• Return a BoxedFilter from the routes() function of each of the 14 route modules, so the router composes 14 type-erased filters instead of a single nested tree of about 200 combinators and closures whose symbol names reached 137,000 characters.
• Leave the router and the 30 individual routes untouched: matching order, rejections, status codes and reply types are unchanged, and the 150 route tests pass as they are.

Pre-submit checklist

• Branch
  • Tests are provided (if possible)
  • Crates versions are updated (if relevant)
  • CHANGELOG file is updated (if relevant)
  • Commit sequence broadly makes sense
  • Key commits have useful messages
• PR
  • All check jobs of the CI have succeeded
  • Self-reviewed the diff
  • Useful pull request description
  • Reviewer requested

Comments

The cost was in rustc's monomorphization collector, not in the linker: every crate that instantiates the router walked the whole type again, while rust-lld relinks the 667 MB debug binary in 2.5 s.

Binary crate after a library edit, dev profile, -Ztime-passes:

Pass	Before	After
total	58.7 s	7.7 s
monomorphization collector walk	48.7 s	1.2 s
link	1.6 s	1.3 s

Measured against main on 6 cores, after a library edit:

Metric	Before	After
binary step, dev	58.5 s	7.4 s
full crate build, dev	1 min 08 s	29 s
debug binary size	667 MB	476 MB

Measured on the earlier baseline branch:

Metric	Before	After
binary step, release	70 s	17 s
unit-test binary compile	122 s	77 s
unit-test binary size	880 MB	562 MB
longest symbol name	137,231 chars	11,965 chars
.debug_str section	361 MB	180 MB
library compile, dev	10 s	20 s
library compile, release	24 s	36 s

• The library's own compile gets slower because each module now instantiates its boxed wrapper once; the net effect is still a large win for every workflow.
• Serving performance is unaffected in practice: boxing adds one allocation and two virtual calls per module attempt, and removes the per-request clone of the whole right-hand subtree that warp's Or performed at every level.
• Integration test binaries are unchanged, as they never instantiate the router.

[github-actions]

Test Results

     5 files  ± 0     221 suites  ±0   59m 10s ⏱️ - 17m 29s
 3 590 tests  - 41   3 590 ✅  - 41  0 💤 ±0  0 ❌ ±0 
11 712 runs   - 47  11 712 ✅  - 47  0 💤 ±0  0 ❌ ±0 

Results for commit 8106e8d. ± Comparison against base commit baaec57.

This pull request removes 41 tests.

mithril-stm ‑ circuits::halo2::tests::golden::cases::negative::slow::index_out_of_bounds
mithril-stm ‑ circuits::halo2::tests::golden::cases::negative::slow::index_too_large_for_circuit_range
mithril-stm ‑ circuits::halo2::tests::golden::cases::negative::slow::indices_not_increasing
mithril-stm ‑ circuits::halo2::tests::golden::cases::negative::slow::leaf_merkle_path_mismatch
mithril-stm ‑ circuits::halo2::tests::golden::cases::negative::slow::leaf_swap_keep_merkle_path
mithril-stm ‑ circuits::halo2::tests::golden::cases::negative::slow::leaf_wrong_verification_key
mithril-stm ‑ circuits::halo2::tests::golden::cases::negative::slow::merkle_path_corrupt_sibling
mithril-stm ‑ circuits::halo2::tests::golden::cases::negative::slow::merkle_path_flip_position
mithril-stm ‑ circuits::halo2::tests::golden::cases::negative::slow::merkle_path_length_long
mithril-stm ‑ circuits::halo2::tests::golden::cases::negative::slow::merkle_path_length_short
…