feat(softdevice): use ConnectionPriority, and add PeripheralLatency

[acouvreur]

This PR is on top of #472. The difference shows only the SoftDevice change. It is independent of #473, which does the same work for Windows.

What this does

The SoftDevice sets the connection parameters in the controller with sd_ble_gap_conn_param_update, thus it uses the explicit fields. This PR fills each field that the caller does not set from ConnectionParams.Priority. Thus code that sets only a priority also gets a result on this platform.

This makes the rule in #472 more exact. That PR says that a platform of this type ignores Priority. This is correct, but it is not sufficient, because a caller that sets only a priority then gets no result. The rule is now that explicit values stay, and that the priority fills the other fields. ConnectionParams.Resolved shows the two together.

The table is in gap.go

ConnectionPriority.Params gives the parameters for each priority.

Priority	Min	Max	Peripheral latency	Supervision timeout
Throughput	15 ms	30 ms	0	2 s
Balanced	30 ms	60 ms	0	4 s
PowerSaving	120 ms	165 ms	4	6 s

The table is in gap.go and not in each platform. Thus all backends of this type agree, and a test can make sure of the values. The test makes sure of two conditions.

• The values obey the guidelines from Apple (section 35.6 Connection Parameters). The intervals are multiples of 15 ms, and the supervision timeouts are between 2 and 6 seconds. https://developer.apple.com/accessories/Accessory-Design-Guidelines.pdf
• The supervision timeout is longer than (1 + peripheral latency) × max interval × 2. The SoftDevice rejects other values.

Please examine the three rows with care. They are my selection, and I did not copy them from a specification. The values for PowerSaving are an estimate of a good compromise.

PeripheralLatency

The table needs a field that ConnectionParams did not have. The difference between power saving and balanced is the number of connection events that the peripheral can skip. The value of ble_gap_conn_params_t.slave_latency was always 0, thus the user could not set it. ConnectionParams.PeripheralLatency now sets it. This is also a correction on its own.

This is also why the enumeration in #422 could not operate. Without this field, ConnectionLatencyHigh had no equivalent on the backends that can use it.

Tests

go build, go vet and go test are satisfactory on the host. golangci-lint finds no new problems. TinyGo builds pca10056-s140v7 and pca10040-s132v6. The second target uses examples/heartrate-monitor, which includes the central file that this PR changes.

I did not test this on real hardware yet. I did not make sure on a board that a PowerSaving request gives the correct interval and latency in a connection parameters update event.