tracing: improve interfaces around process boundaries

[irfansharif]

We can rid ourselves of the opentracing cruft just a bit more, and add
some type-safety while doing so. While here, we'll also add some text
explaining how tracing data gets propagated across process boundaries.

Release note: None

---

First commit is from #59815, ignore here.

[cockroach-teamcity]

This change is 

[irfansharif]

CI's failing due to something else:

panic: geos: error during GEOS init: geos: cannot load GEOS from dir "/go/src/github.com/cockroachdb/cockroach/lib": geos error: /lib/x86_64-linux-gnu/libm.so.6: version `GLIBC_2.29' not found (required by /go/src/github.com/cockroachdb/cockroach/lib/libgeos.so)

[tbg]

Reviewed 22 of 22 files at r1, 14 of 14 files at r2, 9 of 9 files at r3.
Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @angelapwen, @irfansharif, and @knz)

---

pkg/util/tracing/tracer.go, line 643 at r2 (raw file):

//
// A context wrapping the newly created span is returned, along with the span
// itself. If non-nil, the caller is responsible to eventually Finish() it.

nit: for eventually finishing it

---

pkg/util/tracing/tracer.go, line 668 at r2 (raw file):

//
// A context wrapping the newly created span is returned, along with the span
// itself. If non-nil, the caller is responsible to eventually Finish() it.

nit: for eventually finishing it

[irfansharif]

Re-ordered the commits to decouple it from #59815. TFTR!

bors r+

[craig]

Build succeeded:

• GitHub CI (Cockroach)

[RaduBerinde]

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @angelapwen, @irfansharif, @knz, and @tbg)

---

pkg/kv/kvserver/replica_proposal.go, line 854 at r5 (raw file):

	traceCarrier := tracing.MapCarrier{
		Map: make(map[string]string),

[nit] maybe put this into a tracing.MakeMapCarrier

---

pkg/util/tracing/tracer.go, line 637 at r4 (raw file):

// details.
//
// The recordings from these spans will be automatically propagated to the

This deserves some more comment, given that the forked span can outlive the parent. There will be a "partial" recording in the parent, depending when it is retrieved, right?

[nit] s/these spans/the forked span

[RaduBerinde]

---

pkg/util/tracing/tracer.go, line 637 at r4 (raw file):

Previously, RaduBerinde wrote…

This deserves some more comment, given that the forked span can outlive the parent. There will be a "partial" recording in the parent, depending when it is retrieved, right?

[nit] s/these spans/the forked span

Ah, I see there's another PR proposing changes here, never mind