refactor: decompose prune.go by protocol

[ShocOne]

internal/sdkbind/prune.go sat at 799 lines against an 800-line ceiling, so
the next change to it failed scripts/repo_hygiene_gate.sh. handoff.md:251
names it as blocking work: the largest attribute-level refusal family left —
234 carries no Get… to read — lives in this package.

The split

Three files, by protocol rather than by size:

File	Lines	Contents
prune.go	352	Prune, pruner, resolveType, remove, the four per-kind entities (resource, datasource, listResource, action), settleUpdateBody, copyFieldBindings, unbuildableReason
prune_calls.go	328	Resolving a drafted call against the SDK that has to make it: resolveCall, resolveFieldHop, repairBodySetter, settleCall, writeModelFor, settleParamTypes, repairIndexer, recordTypePackage
prune_list.go	139	Settling how a list call reaches its elements: resolveListElement, sliceGetters, sliceFields, pickByName

The two new files carry a header saying what the group is for — a collection
comes back as a slice, as an envelope with a getter, or as an envelope with a
field, and which of those it is decides what the generated iteration reads.

Measured effect

None, and that is the acceptance test.

unsupported.json is byte-identical on all three pilots — verified with
diff, not by comparing totals:

github         unsupported.json byte-identical
jamfpro        unsupported.json byte-identical
thousandeyes   unsupported.json byte-identical

tfpfgen provider verify reports no drift on all three (3358 / 2935 / 1420
files), and postcheck passes.

No function was renamed, no signature changed, and no comment was rewritten —
the declarations moved and nothing else, so the diff is reviewable as a move.

Gates

• gofmt -l internal cmd — empty
• scripts/repo_hygiene_gate.sh — pass (this is what it was failing)
• golangci-lint run — 0 issues
• go build ./... && go vet ./... — pass
• go test ./... — pass, internal/sdkbind 85.0%, 91.4% total

🤖 Generated with Claude Code