Let $ (C_i)_{i \in I} $ be a chain of subsets of $ \mathbb{N} $ where $ I $ is an arbitrary indexing set. Does there always exist a countable subchain $ (C_{i_n})_{n \in \mathbb{N}} $ such that $ \bigcap_{i \in I} C_i = \bigcap_{n \in \mathbb{N}} C_{i_n} $?
This seems like a fairly basic result, but I haven't found it anywhere.
I have a proof in mind using choice. Let $ J = \bigcap_{i \in I} C_i $. We can enumerate $ J^c = \{a_1, a_2, \dots \} $. Then, well-order $ I $. We proceed recursively as follows. For the base case, let $ i_1 $ be the minimal index in the well-order such that $ a_1 \not \in C_{i_1} $. Then, assume $ i_1, \dots, i_n $ have been defined so that $ C_{i_k} \supset C_{i_{k + 1}} $ and $ a_k \not \in C_{i_k} $ for $ 1 \leq k \leq n $. We choose $ i_{n + 1} $ to be the minimal index such that $ C_{i_n} \supset C_{i_{n + 1}} $ and $ a_{n + 1} \not \in C_{i_{n + 1}} $.
As $ (C_{i_n})_{n \in \mathbb{N}} $ is a subchain of $ (C_i)_{i \in I} $, $ \bigcap_{n \in \mathbb{n}} C_{i_n} \supseteq \bigcap_{i \in I} C_i $. Moreover, for any $ a \not \in \bigcap_{i \in I} C_i $, $ a \not \in \bigcap_{n \in \mathbb{n}} C_{i_n} $ by construction, so we have equality.
Is this correct? If so, can it be done without choice?
Your argument looks correct!
As to your second question, yes, choice is in fact necessary. To see this I think it's helpful to switch from chains of subsets of $\mathbb{N}$ to chains of subsets of $\mathbb{Q}$ (since there is a bijection between those two sets, this is a benign modification). We can now associate, to each set $A$ of real numbers, a chain $\mathcal{C}_A$ of sets of rational numbers via Dedekind cuts: $$\mathcal{C}_A=\{\{q\in\mathbb{Q}: q<a\}: a\in A\}.$$ Now it turns out that it is consistent with $\mathsf{ZF}$ (= set theory without choice) that there is an uncountable set of reals $X$ with no countably infinite subset. In particular, this means that $\mathcal{C}_X$ has no countably infinite chains at all, let alone countably infinite chains with the same intersection as $\bigcap \mathcal{C}_X$, and so we're done. (OK fine technically we also need that $X$ have no least element, but this is easy to ensure: if $X$ is as above then the set of elements of $X$ with only finitely many predecessors in $X$ is itself finite, and we can WLOG remove this finite set from $X$.)
However, this $\mathcal{C}_X$ does "almost" have a countable subchain with the same intersection: there is a countable set of reals $Y$ such that $$\bigcap \mathcal{C}_{X\cup Y}=\bigcap \mathcal{C}_X=\bigcap \mathcal{C}_Y.$$ Just take $Y$ to consist of the countably many rationals greater than some element of $X$. Now most of the elements of $\mathcal{C}_Y$ won't be elements of $\mathcal{C}_X$, but in a sense adding them doesn't disrupt things too much.
This raises the question of whether choice is needed for the following seemingly-weaker result:
In fact choice is not needed here since we can modify your argument to apply choicelessly to $\mathcal{C}^+$ as follows: