[Abhishek-B-R]

Motivation:
etype.MarshalTime formats with time.RFC3339, which has no fractional
seconds. The generated code uses it for path, query and header values
(genutil.MarshalBuiltin and MarshalQueryOrHeader), so a time.Time
passed to another service in one of those places, or returned in a
response header, is silently truncated to the second. The same value in
a JSON body keeps full precision, so the behaviour depends on where the
field happens to live.

This bites things like keyset pagination: a CreatedAfter time.Time
query param holding a Postgres timestamp (microseconds) arrives as the
whole second, and the callee returns rows the caller has already seen.

Before: 2026-10-28T09:30:00.123456Z is sent as 2026-10-28T09:30:00Z.
After: it is sent unchanged.

Approach:
Format with time.RFC3339Nano. It trims trailing zeros, so whole-second
values encode exactly as they did before, and UnmarshalTime already
parses fractional seconds (Go's RFC3339 parser accepts them), so old and
new runtimes on either side of a call still understand each other.

Validation:
Added TestMarshalTime in runtimes/go/appruntime/shared/etype, covering
whole seconds, microseconds, nanoseconds and a +05:30 offset, each checked
for the exact string and for a lossless round trip through UnmarshalTime.
On main the three sub-second cases fail; with the change all pass.

• cd runtimes/go && go test -short -tags=dev_build ./appruntime/shared/etype/ (also with TZ=Asia/Kolkata)
• cd runtimes/go && go test -short -tags=dev_build ./... passes
• go vet and gofmt -l clean on the package

The generated Go client (internal/gocodegen/marshalling.go, FromTime)
has the same time.RFC3339 format. I left it out to keep this to the
runtime, since changing it means regenerating the client goldens. Happy
to do it here or in a follow-up if you want it.

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}