We should allow a developer writing runtime benchmarks to whitelist a DB key as a part of their benchmarking code.

This will most commonly be used to whitelist the caller of the benchmark extrinsic. This would mean we could use accounts other than caller, 0, 0 as the caller of a function, and correctly ignore changes to that account.

This would involve adding/change the host functions exposed by the benchmarking DB. Right now we have set_whitelist which allows you to set a complete whitelist (or reset a whitelist by setting it to an empty vec). We should probably introduce an add_to_whitelist function which simply appends new keys to the existing whitelist.