Turn Expr into a proper range instead of a view

[Krzmbrzl]

Previously, the Expr implementation used range-v3's view facade to
implement range semantics on Expr objects. However, views are different
things from containers which had some more or less subtle consequences.
See also https://stackoverflow.com/a/31462435

This PR replaces the view facade with virtual begin/end
implementations returning ExprIterator objects. Hence, Expr is now a
fully fledged container that is usable as a random-access range.

[Krzmbrzl]

Okay, I'll have to change the impl to make some protected helper functions like get_begin() & Co which are then called in Expr's base begin()/end() functions to get rid of these ridiculous shadowing issues.

[evaleev]

Post-merge review of this PR turned up three defects; #594 fixes them, with a regression test for each.

1. The const/non-const iterator overloads never worked. ExprIteratorImpl's heterogeneous operator-/operator==/operator<=> read other.ptr_ of the other specialization, which is private and has no friend declaration — a hard error the moment any of them is instantiated. Nothing in tree mixes the two types, which is why CI is green, but expr.begin() != expr.cend() does not compile, and neither does pairing sequant::cbegin(ExprPtr const&) with the non-const sequant::end(ExprPtr&). There is also no ExprIterator → ConstExprIterator conversion. (Also removed operator-(difference_type, ExprIteratorImpl): n - it is not a valid random-access-iterator expression and it silently computed it - n.)

2. Expr::at() lost its bounds check. The ranges::view_interface::at() that went away used to throw; the replacement forwards to operator[], which is guarded only by SEQUANT_ASSERT. With SEQUANT_ASSERT_BEHAVIOR=IGNORE, sum.at(p) at optimize/sum.cpp:118 degrades to an out-of-bounds read. back() is worse: at(size() - 1) on any atom computes at(SIZE_MAX).

3. Product::end_subexpr() stopped invalidating the memoized hash. The old Product::end_cursor() called reset_hash_value(). Without it, *(--product.end()) = new_factor leaves a stale hash — tripping the assert in Product::memoizing_hash(), or silently making static_equal() return false for equal products. (Conversely, Sum::begin_subexpr() gained a reset that Sum::begin_cursor() never had — that one is correct, and Follow-ups to #593: iterator interop, checked Expr::at(), hash invalidation #594 extends it to Sum::end_subexpr() for symmetry.)

#594 also moves begin/end/size/empty/operator[]/at/front/back back into the header; they are one-liners on the hottest paths in the library (Expr::is_atom() is called from visit_impl(), is_scalar(), ExprRange::next_atom(), Wick and canonicalization) and out-of-line definitions make each a non-inlinable cross-TU call on top of the virtual dispatch.