Count a dirty epoch on functions and on the type database ##analysis

[0verflowme]

Anything that caches work derived from a function's typed metadata, or from the type database, needs to know when that input changed. has_changed is a flag that one consumer clears, so a second consumer never sees the change. An epoch is a counter that only moves forward, and every consumer keeps the value it last saw.

• RAnalFunction.dirty_epoch, bumped by r_anal_function_bump_dirty_epoch, on: relocate, rename, variable set/rename/retype, signature set, calling convention change, afB, afS, and when r_anal_function_cc resolves a lazy dyncc marker. Zero is never a valid value, so a consumer that starts from zero always sees the first change.
• RAnal.type_dirty_epoch, bumped by r_anal_types_bump_dirty_epoch, on: type database loads and reloads, parsed type save and remove, base type save, signature delete, and type links. The link writers (r_type_set_link, r_type_link_offset, r_type_unlink) do not see RAnal, so r_anal_types_set_link, r_anal_types_set_link_offset and r_anal_types_unlink wrap them and the three core callers go through the wrappers.
• r_anal_function_set_callconv validates the convention and bumps; afc uses it instead of writing fcn->callconv directly.

The function bump also sets has_changed, so existing readers of that flag keep working. test_anal_function checks that relocate and rename move the epoch.

Split out of #26701.