fix: a filter survives only where the field it selects on does

[ShocOne]

Two defects in #105, both found by re-emitting Jamf Pro, ThousandEyes and
GitHub rather than by any test. The fictional fixtures produce neither shape.

A filter outlived the field it selects on

Filters are derived from the listed object in derive.go, which runs before
sdkbind prunes. Binding deletes what the SDK cannot carry, and the item model
is built from what survives — so a filter over a deleted field compared against
a model field that no longer existed:

internal/services/datasources/account_groups/v1/account_group/read.go:39:70:
  item.AccountToken undefined (type *AccountGroupItemModel has no field or method AccountToken)

Emission now reconciles the two, keeping only the filters whose field survived.
Derivation still declares the intent; emission is the only place that knows
what binding left.

An integer id was declared as a string

The id short-circuit hard-coded data.ID.ValueString(), which was correct only
while the argument was the always-string filter_value. A filter takes the type
of the field it selects on, and Jamf Pro keys api_integration by an integer:

internal/services/datasources/api_integrations/v1/api_integration/read.go:59:17:
  data.ID.ValueString undefined (type types.Int64 has no field or method ValueString)

The parameter is built through buildCallPlan now — the same path the resource
read and the lookup datasource already use, which converts the attribute to the
parameter's type and declares any import that needs. readPlanWithoutParams and
the hand-written IDParamDecl are deleted; there is nothing left that assumed
the id was a string.

Re-measured

All three trees regenerate and pass postcheck — go mod tidy, go build,
go vet:

Document	Resources	Data sources	List resources	Actions
Jamf Pro	76	211	39	101
GitHub	59	323	23	77
ThousandEyes	35	95	11	51
Total	170	629	73	229

README.md carried 161 / 272 / 69 / 210. Actions moving to 229 is the
lone-write change in #104; the rest predates this work and was simply stale, so
the line is re-measured rather than adjusted.

Verification

Added a regression test that deletes one field binding and proves the filter
goes with it — out of the schema, the model and the match — while the filters
whose fields survived are untouched. make check passes at 91.3%.

The load-bearing check remains generating the three documents, which is what
caught both of these. Neither would have been found by the fixtures.

🤖 Generated with Claude Code