Expose hov:lanes and hov:minimum as raw text

[dsaderholm]

Closes #130. Entirely reasonable to close this instead if it is not worth the maintenance — the coverage numbers below are regional and I said in the issue I would put it behind the overtaking work.

What it does

Passes hov:lanes and hov:minimum through untouched, in the same shape as maxSpeedConditional: read in the generator, stored on the offline Way, memoized on maps.Way, published on MapdOut. No lane indexing and no occupancy logic in mapd — whether an HOV lane may be used depends on occupancy, tolling and local rules, so consumers evaluate the tag themselves.

Two fields, not the directional variants

From taginfo:

hov:lanes                19,596
hov:minimum              12,215
hov:lanes:conditional     1,207
hov:lanes:backward          252
hov:lanes:forward           216

The forward/backward forms are about 1% of hov:lanes, so carrying them would cost every tile two more Text fields for almost nothing. Left them out, and hov:lanes:conditional too. Happy to add them if you would rather be symmetric with the maxspeed handling.

Format and coverage

Where the tag exists the HOV lane is the leftmost one — designated is the first entry on all 573 tagged ways I sampled on I-15 in Utah, and the pipe count always matches lanes.

I-15 Utah              1269 ways    45%
I-15 Arizona/Nevada     500 ways    22%
I-15 California         544 ways     0%

The California zero is not missing data: there the HOV lane is mapped as its own carriageway, so the matched way genuinely has no hov:lanes and the field is correctly empty. docs/outputs.md says so, since it would otherwise look like a bug.

Compatibility

Both fields are appended at the end of their structs, so tiles written before this read empty rather than failing. MapdOut's pointer count goes 6 → 8, which is the normal additive change.

Verification, including what I could not run

capnp 1.3.0 with capnpc-go v3.1.0-alpha.1. Before making any change I regenerated both schemas unmodified and confirmed a zero diff against the committed .capnp.go, so the generated output here is byte-comparable with yours rather than reformatted by a different toolchain version. The generated diff is the two accessor sets, the pointer count, and the embedded schema blob.

go build, go vet and go test were not run locally — gomsgq needs linux and cgo, and I wrote this on Windows. I verified by inspection that every generated symbol the hand-written code calls exists with the expected signature, and that the state.go call is the same shape as the SetRoadName line above it. CI covers the rest, and I will fix anything it finds.

I did not run gofumpt -w: with the current release it wants to rewrite files this PR does not touch, and I did not want to bury the change.

No test added, matching the existing conditional-tag accessors, which are also untested — the parser has tests because there is something to parse, and the point of this one is that there is not.

Happy to test a build on device.

🤖 Generated with Claude Code