feat(control_plane): Phase 1b -- depend on asap-ir for real, not a copy

[zzylol]

Summary

Supersedes #391. Please close #391 in favor of this one -- Phase 1's
approach (copy ASAPController's AggIntent vocabulary into
control_plane's own agg_intent.rs) turned out not to be a merge at
all: it's a second, independently-maintained copy that drifts again the
moment either side changes anything, which is exactly the "two-repo
split forces coordinated PRs" problem design-controller-into-backend.md
already flagged. This PR adds a real git dependency on ASAPController's
asap-ir crate (pinned to c73c2a5, ASAPController#137's merge commit --
no tagged releases exist yet) and deletes the local AggIntent
definition.

What only showed up once depending on the real type

A local copy can freely grow extra fields; a dependency on a type you
don't own can't. Two coupling issues surfaced that Phase 1 had no way to
see:

• Column/DataType/AccuracyTarget. AggIntent's own methods are
  typed against asap_ir's versions of these, not control_plane's local
  ones. ColumnId is a bare usize alias on both sides (free). For
  Column/DataType I convert at the agg_intent.rs boundary rather
  than merging schema.rs wholesale -- that cascades into Schema/
  QueryExpr (~38 Column{} literals across the repo), which is Phase 2
  scope, not this PR's. For AccuracyTarget I went the other way and
  swapped control_plane's local type for asap_ir's directly (types_v2:: AccuracyTarget is now pub use asap_ir::types::AccuracyTarget) --
  checked first that no external YAML/JSON persists the old wire shape
  (#[serde(tag="kind", content="value")] with an eps field vs.
  asap_ir's default tagging with epsilon); only one in-Rust test
  fixture needed updating, so converting at ~400 call sites just to keep
  a differently-shaped local copy wasn't worth it.
• Orphan rules. .archive_only() / .output_column() (control_plane
  concepts, not shared ones) can't be inherent methods on a foreign type
  anymore -- they're free functions now (archive_only(&intent),
  output_column(&intent, &col)). asap_ir's own methods
  (.input_col(), .is_per_series()) keep method syntax.

Frequency (control_plane's point-frequency-via-CMS query -- see
ASAPController#137 for why it's not the same thing as
RankingMeasure::Frequency) is now carried as AggIntent::Extension { ext_kind: "frequency", payload } via frequency()/as_frequency()
helpers.

window: Duration on Rate/Increase/Changes/Delta/IDelta/
Deriv/Resets -- Phase 1 deferred this by just adding the field to its
local copy -- is gone for real; you can't add a field to a variant you
don't own. I grepped for actual reads of the field (not constructions)
before assuming this needed a big restructure: there was exactly one
real consumer (bind_exact_agg.rs's zero-window rejection guard), which
now reads the window off the enclosing QueryExpr::Window node it
already has in scope.

Also needed

.cargo/config.toml (net.git-fetch-with-cli = true) at the workspace
root -- Cargo's bundled libgit2 doesn't use the configured SSH agent the
way the system git CLI does. data_plane already carries this same
setting locally for its own git path/deps; this just makes it apply
workspace-wide so control_plane's new git dependency resolves too.

Verification

• cargo build --workspace -- clean
• cargo test -p control_plane --lib -- 828 passed, 1 pre-existing
  failure (invalid_sketch_type_override_falls_back_to_default),
  confirmed identical on unmodified main -- not caused by this change
  (same failure feat(control_plane): Phase 1 -- merge AggIntent vocabulary from ASAPController #391 already flagged)
• rustfmt applied to all changed files

Still flagging from #391: analyzer-parity-matrix.md's named acceptance
test doesn't exist in current data_plane under that name -- Phase 3
needs to locate or rebuild it before starting.

Test plan

• cargo build --workspace
• cargo test -p control_plane --lib
• Reviewer sign-off, then close feat(control_plane): Phase 1 -- merge AggIntent vocabulary from ASAPController #391

🤖 Generated with Claude Code